[{"layout_dets":[{"category_id":0,"poly":[282.1632080078125,156.2249755859375,1416.6795654296875,156.2249755859375,1416.6795654296875,313.81280517578125,282.1632080078125,313.81280517578125],"score":0.999998927116394},{"category_id":1,"poly":[861.656982421875,522.7763061523438,1569.3853759765625,522.7763061523438,1569.3853759765625,656.883544921875,861.656982421875,656.883544921875],"score":0.9999970197677612},{"category_id":1,"poly":[131.8020782470703,924.7362670898438,838.9530639648438,924.7362670898438,838.9530639648438,1323.7529296875,131.8020782470703,1323.7529296875],"score":0.9999949932098389},{"category_id":1,"poly":[133.32005310058594,1324.5035400390625,839.2289428710938,1324.5035400390625,839.2289428710938,1589.4503173828125,133.32005310058594,1589.4503173828125],"score":0.999994158744812},{"category_id":1,"poly":[863.3811645507812,1486.610107421875,1569.2880859375,1486.610107421875,1569.2880859375,1852.443603515625,863.3811645507812,1852.443603515625],"score":0.9999936819076538},{"category_id":1,"poly":[862.9096069335938,1187.8067626953125,1568.2279052734375,1187.8067626953125,1568.2279052734375,1486.08935546875,862.9096069335938,1486.08935546875],"score":0.9999932050704956},{"category_id":1,"poly":[131.8186492919922,1652.7752685546875,837.5543823242188,1652.7752685546875,837.5543823242188,2019.429443359375,131.8186492919922,2019.429443359375],"score":0.9999901056289673},{"category_id":0,"poly":[375.1526794433594,881.8807983398438,594.3075561523438,881.8807983398438,594.3075561523438,913.4786987304688,375.1526794433594,913.4786987304688],"score":0.9999892115592957},{"category_id":2,"poly":[636.1867065429688,2099.795654296875,1063.7423095703125,2099.795654296875,1063.7423095703125,2124.524169921875,636.1867065429688,2124.524169921875],"score":0.9999860525131226},{"category_id":0,"poly":[375.91864013671875,1610.209228515625,592.8395385742188,1610.209228515625,592.8395385742188,1641.5789794921875,375.91864013671875,1641.5789794921875],"score":0.9999815821647644},{"category_id":4,"poly":[860.6583251953125,995.6574096679688,1569.622314453125,995.6574096679688,1569.622314453125,1126.8409423828125,860.6583251953125,1126.8409423828125],"score":0.9999815821647644},{"category_id":1,"poly":[443.1008605957031,353.8008728027344,1250.531494140625,353.8008728027344,1250.531494140625,464.65576171875,443.1008605957031,464.65576171875],"score":0.9999791979789734},{"category_id":1,"poly":[130.8282928466797,523.2079467773438,836.5639038085938,523.2079467773438,836.5639038085938,862.0206909179688,130.8282928466797,862.0206909179688],"score":0.9999784231185913},{"category_id":1,"poly":[862.6514282226562,1851.426513671875,1568.510498046875,1851.426513671875,1568.510498046875,2017.93359375,862.6514282226562,2017.93359375],"score":0.9999769926071167},{"category_id":3,"poly":[882.3795166015625,685.376708984375,1544.4088134765625,685.376708984375,1544.4088134765625,969.22265625,882.3795166015625,969.22265625],"score":0.9994785785675049},{"category_id":13,"poly":[1195,1062,1226,1062,1226,1096,1195,1096],"score":0.88,"latex":"d_{p}"},{"category_id":13,"poly":[1304,1030,1327,1030,1327,1061,1304,1061],"score":0.65,"latex":"\\bar{\\bf p}"},{"category_id":15,"poly":[344.0,165.0,1354.0,172.0,1353.0,236.0,344.0,229.0],"score":0.99,"text":"Real-time Temporal Stereo Matching"},{"category_id":15,"poly":[293.0,254.0,1402.0,254.0,1402.0,309.0,293.0,309.0],"score":0.99,"text":"using Iterative Adaptive Support Weights"},{"category_id":15,"poly":[864.0,527.0,1568.0,527.0,1568.0,559.0,864.0,559.0],"score":0.99,"text":"disparity map. Note that individual disparities can be converted"},{"category_id":15,"poly":[864.0,561.0,1568.0,561.0,1568.0,594.0,864.0,594.0],"score":0.98,"text":"to actual depths if the geometry of the camera setup is"},{"category_id":15,"poly":[859.0,587.0,1568.0,591.0,1568.0,630.0,859.0,626.0],"score":0.98,"text":" known, i.e., the stereo configuration of cameras has been pre-"},{"category_id":15,"poly":[862.0,626.0,984.0,626.0,984.0,658.0,862.0,658.0],"score":1.0,"text":"calibrated."},{"category_id":15,"poly":[155.0,921.0,839.0,924.0,838.0,963.0,155.0,960.0],"score":0.98,"text":" Modern stereo matching algorithms achieve excellent results"},{"category_id":15,"poly":[127.0,956.0,838.0,958.0,838.0,997.0,127.0,995.0],"score":0.98,"text":" on static stereo images, as demonstrated by the Middlebury"},{"category_id":15,"poly":[132.0,995.0,836.0,995.0,836.0,1027.0,132.0,1027.0],"score":0.98,"text":"stereo performance benchmark [1], [2]. However, their ap-"},{"category_id":15,"poly":[134.0,1027.0,834.0,1027.0,834.0,1059.0,134.0,1059.0],"score":1.0,"text":"plication to stereo video sequences does not guarantee inter-"},{"category_id":15,"poly":[134.0,1061.0,836.0,1061.0,836.0,1093.0,134.0,1093.0],"score":0.99,"text":"frame consistency of matches extracted from subsequent stereo"},{"category_id":15,"poly":[132.0,1095.0,838.0,1095.0,838.0,1125.0,132.0,1125.0],"score":0.99,"text":"frame pairs. The lack of temporal consistency of matches"},{"category_id":15,"poly":[134.0,1128.0,836.0,1128.0,836.0,1157.0,134.0,1157.0],"score":1.0,"text":"between successive frames introduces spurious artifacts in the"},{"category_id":15,"poly":[132.0,1160.0,836.0,1160.0,836.0,1192.0,132.0,1192.0],"score":0.99,"text":"resulting disparity maps. The problem of obtaining temporally"},{"category_id":15,"poly":[132.0,1194.0,838.0,1194.0,838.0,1226.0,132.0,1226.0],"score":0.98,"text":"consistent sequences of disparity maps from video streams is"},{"category_id":15,"poly":[134.0,1228.0,838.0,1228.0,838.0,1260.0,134.0,1260.0],"score":0.98,"text":"known as the temporal stereo correspondence problem, yet"},{"category_id":15,"poly":[129.0,1258.0,841.0,1260.0,841.0,1293.0,129.0,1290.0],"score":0.98,"text":"the amount of research efforts oriented towards finding an"},{"category_id":15,"poly":[134.0,1292.0,760.0,1292.0,760.0,1325.0,134.0,1325.0],"score":0.99,"text":"effective solution to this problem is surprisingly small."},{"category_id":15,"poly":[157.0,1320.0,836.0,1322.0,836.0,1361.0,157.0,1359.0],"score":0.98,"text":" A method is proposed for real-time temporal stereo match-"},{"category_id":15,"poly":[134.0,1361.0,836.0,1361.0,836.0,1393.0,134.0,1393.0],"score":1.0,"text":"ing that efficiently propagates matching cost information be-"},{"category_id":15,"poly":[134.0,1393.0,836.0,1393.0,836.0,1425.0,134.0,1425.0],"score":0.99,"text":"tween consecutive frames of a stereo video sequence. This"},{"category_id":15,"poly":[132.0,1423.0,834.0,1425.0,834.0,1458.0,132.0,1455.0],"score":0.98,"text":"method is invariant to the number of prior frames being"},{"category_id":15,"poly":[134.0,1458.0,836.0,1458.0,836.0,1490.0,134.0,1490.0],"score":0.99,"text":"considered, and can be easily incorporated into any local stereo"},{"category_id":15,"poly":[132.0,1492.0,836.0,1492.0,836.0,1524.0,132.0,1524.0],"score":0.98,"text":"method based on edge-aware filters. The iterative adaptive"},{"category_id":15,"poly":[132.0,1526.0,838.0,1526.0,838.0,1558.0,132.0,1558.0],"score":0.99,"text":"support matching algorithm presented in [3] serves as a"},{"category_id":15,"poly":[132.0,1558.0,557.0,1558.0,557.0,1590.0,132.0,1590.0],"score":0.99,"text":"foundation for the proposed method."},{"category_id":15,"poly":[887.0,1483.0,1571.0,1485.0,1571.0,1524.0,887.0,1522.0],"score":0.98,"text":" In contrast, local methods, which are typically built upon"},{"category_id":15,"poly":[859.0,1517.0,1573.0,1519.0,1573.0,1558.0,859.0,1556.0],"score":0.97,"text":" the Winner-Takes-All (WTA) framework, have the property of "},{"category_id":15,"poly":[864.0,1556.0,1566.0,1556.0,1566.0,1588.0,864.0,1588.0],"score":0.99,"text":"computational regularity and are thus suitable for implemen-"},{"category_id":15,"poly":[862.0,1588.0,1566.0,1588.0,1566.0,1620.0,862.0,1620.0],"score":1.0,"text":"tation on parallel graphics hardware. Within the WTA frame-"},{"category_id":15,"poly":[862.0,1616.0,1568.0,1618.0,1568.0,1657.0,862.0,1655.0],"score":0.98,"text":"work, local stereo algorithms consider a range of disparity"},{"category_id":15,"poly":[864.0,1655.0,1566.0,1655.0,1566.0,1687.0,864.0,1687.0],"score":0.98,"text":"hypotheses and compute a volume of pixel-wise dissimilarity"},{"category_id":15,"poly":[862.0,1689.0,1571.0,1689.0,1571.0,1721.0,862.0,1721.0],"score":0.99,"text":"metrics between the reference image and the matched image at"},{"category_id":15,"poly":[862.0,1723.0,1568.0,1721.0,1568.0,1753.0,862.0,1755.0],"score":0.99,"text":"every considered disparity value. Final disparities are chosen"},{"category_id":15,"poly":[864.0,1755.0,1568.0,1755.0,1568.0,1785.0,864.0,1785.0],"score":1.0,"text":"from the cost volume by traversing through its values and"},{"category_id":15,"poly":[866.0,1788.0,1568.0,1788.0,1568.0,1820.0,866.0,1820.0],"score":0.99,"text":"selecting the disparities associated with minimum matching"},{"category_id":15,"poly":[859.0,1817.0,1377.0,1820.0,1377.0,1859.0,859.0,1856.0],"score":0.98,"text":" costs for every pixel of the reference image."},{"category_id":15,"poly":[885.0,1187.0,1571.0,1187.0,1571.0,1226.0,885.0,1226.0],"score":0.97,"text":" In their excellent taxonomy paper [1], Scharstein and"},{"category_id":15,"poly":[864.0,1224.0,1566.0,1224.0,1566.0,1254.0,864.0,1254.0],"score":0.99,"text":"Szeliski classify stereo algorithms as local or global meth-"},{"category_id":15,"poly":[859.0,1249.0,1571.0,1254.0,1570.0,1293.0,859.0,1288.0],"score":0.99,"text":" ods. Global methods, which offer outstanding accuracy, are"},{"category_id":15,"poly":[862.0,1288.0,1571.0,1288.0,1571.0,1327.0,862.0,1327.0],"score":0.98,"text":"typically derived from an energy minimization framework"},{"category_id":15,"poly":[859.0,1322.0,1566.0,1322.0,1566.0,1352.0,859.0,1352.0],"score":0.99,"text":"that allows for explicit integration of disparity smoothness"},{"category_id":15,"poly":[864.0,1357.0,1568.0,1357.0,1568.0,1389.0,864.0,1389.0],"score":0.99,"text":"constraints and thus is capable of regularizing the solution"},{"category_id":15,"poly":[864.0,1391.0,1568.0,1391.0,1568.0,1421.0,864.0,1421.0],"score":1.0,"text":"in weakly textured areas. The minimization, however, is often"},{"category_id":15,"poly":[864.0,1423.0,1568.0,1423.0,1568.0,1455.0,864.0,1455.0],"score":0.99,"text":"achieved using iterative methods or graph cuts, which do not"},{"category_id":15,"poly":[864.0,1458.0,1418.0,1458.0,1418.0,1487.0,864.0,1487.0],"score":0.99,"text":"lend themselves well to parallel implementation."},{"category_id":15,"poly":[155.0,1650.0,839.0,1652.0,838.0,1691.0,155.0,1689.0],"score":0.97,"text":" Stereo matching is the process of identifying correspon-"},{"category_id":15,"poly":[134.0,1687.0,838.0,1687.0,838.0,1719.0,134.0,1719.0],"score":0.99,"text":"dences between pixels in stereo images obtained using a"},{"category_id":15,"poly":[132.0,1723.0,838.0,1721.0,838.0,1753.0,132.0,1755.0],"score":0.98,"text":"pair of synchronized cameras. These correspondences are"},{"category_id":15,"poly":[134.0,1755.0,836.0,1755.0,836.0,1788.0,134.0,1788.0],"score":0.99,"text":"conveniently represented using the notion of disparity, i.e. the"},{"category_id":15,"poly":[134.0,1788.0,836.0,1788.0,836.0,1820.0,134.0,1820.0],"score":1.0,"text":"positional offset between two matching pixels. It is assumed"},{"category_id":15,"poly":[134.0,1822.0,836.0,1822.0,836.0,1854.0,134.0,1854.0],"score":0.99,"text":"that the stereo images are rectified, such that matching pixels"},{"category_id":15,"poly":[132.0,1854.0,836.0,1854.0,836.0,1886.0,132.0,1886.0],"score":0.99,"text":"are confined within corresponding rows of the images and"},{"category_id":15,"poly":[134.0,1888.0,838.0,1888.0,838.0,1918.0,134.0,1918.0],"score":1.0,"text":"thus disparities are restricted to the horizontal dimension, as"},{"category_id":15,"poly":[134.0,1920.0,838.0,1920.0,838.0,1952.0,134.0,1952.0],"score":1.0,"text":"illustrated in Figure 1. For visualization purposes, disparities"},{"category_id":15,"poly":[134.0,1955.0,838.0,1955.0,838.0,1987.0,134.0,1987.0],"score":0.99,"text":"recovered for every pixel of a reference image are stored"},{"category_id":15,"poly":[129.0,1985.0,841.0,1982.0,841.0,2021.0,129.0,2024.0],"score":0.98,"text":"together in the form of an image, which is known as the"},{"category_id":15,"poly":[370.0,885.0,594.0,885.0,594.0,917.0,370.0,917.0],"score":1.0,"text":"1. INTRODUCTION"},{"category_id":15,"poly":[638.0,2099.0,1062.0,2099.0,1062.0,2131.0,638.0,2131.0],"score":0.98,"text":"978-1-4673-5208-6/13/$31.00 @2013 IEEE"},{"category_id":15,"poly":[374.0,1613.0,591.0,1613.0,591.0,1645.0,374.0,1645.0],"score":0.95,"text":"II. BACKGROUND"},{"category_id":15,"poly":[859.0,992.0,1571.0,995.0,1571.0,1034.0,859.0,1031.0],"score":0.99,"text":" Figure 1: Geometry of two horizontally aligned views where p"},{"category_id":15,"poly":[864.0,1098.0,1291.0,1098.0,1291.0,1130.0,864.0,1130.0],"score":0.99,"text":"them along the horizontal dimension."},{"category_id":15,"poly":[859.0,1061.0,1194.0,1059.0,1194.0,1098.0,859.0,1100.0],"score":0.98,"text":" pixel in the target frame, and"},{"category_id":15,"poly":[1227.0,1061.0,1571.0,1059.0,1571.0,1098.0,1227.0,1100.0],"score":0.97,"text":" denotes the disparity between"},{"category_id":15,"poly":[864.0,1034.0,1303.0,1034.0,1303.0,1063.0,864.0,1063.0],"score":0.99,"text":"denotes a pixel in the reference frame,"},{"category_id":15,"poly":[1328.0,1034.0,1566.0,1034.0,1566.0,1063.0,1328.0,1063.0],"score":0.96,"text":" denotes its matching"},{"category_id":15,"poly":[508.0,357.0,1194.0,360.0,1194.0,392.0,508.0,390.0],"score":0.98,"text":"Jedrzej Kowalczuk, Eric T. Psota, and Lance C. P\u00e9rez"},{"category_id":15,"poly":[443.0,392.0,1245.0,392.0,1245.0,424.0,443.0,424.0],"score":0.99,"text":"Department of Electrical Engineering, University of Nebraska-Lincoln"},{"category_id":15,"poly":[614.0,435.0,1081.0,435.0,1081.0,465.0,614.0,465.0],"score":0.99,"text":"[jkowalczuk2,epsota,lperez] @unl.edu"},{"category_id":15,"poly":[159.0,527.0,836.0,527.0,836.0,559.0,159.0,559.0],"score":0.98,"text":"Abstract-Stereo matching algorithms are nearly always de-"},{"category_id":15,"poly":[132.0,555.0,838.0,555.0,838.0,587.0,132.0,587.0],"score":0.98,"text":"signed to find matches between a single pair of images. A method"},{"category_id":15,"poly":[134.0,580.0,836.0,580.0,836.0,612.0,134.0,612.0],"score":1.0,"text":"is presented that was specifically designed to operate on sequences"},{"category_id":15,"poly":[132.0,605.0,838.0,607.0,838.0,646.0,132.0,644.0],"score":0.99,"text":"of images. This method considers the cost of matching image"},{"category_id":15,"poly":[132.0,637.0,838.0,637.0,838.0,669.0,132.0,669.0],"score":0.98,"text":"points in both the spatial and temporal domain. To maintain"},{"category_id":15,"poly":[134.0,667.0,838.0,667.0,838.0,699.0,134.0,699.0],"score":0.97,"text":"real-time operation, a temporal cost aggregation method is used"},{"category_id":15,"poly":[132.0,692.0,836.0,692.0,836.0,722.0,132.0,722.0],"score":0.98,"text":"to evaluate the likelihood of matches that is invariant with respect"},{"category_id":15,"poly":[127.0,717.0,841.0,715.0,841.0,754.0,127.0,756.0],"score":0.97,"text":"to the number of prior images being considered. This method"},{"category_id":15,"poly":[127.0,742.0,841.0,745.0,841.0,784.0,127.0,781.0],"score":0.98,"text":"has been implemented on massively parallel GPU hardware,"},{"category_id":15,"poly":[132.0,777.0,838.0,777.0,838.0,809.0,132.0,809.0],"score":0.99,"text":"and the implementation ranks as one of the fastest and most"},{"category_id":15,"poly":[132.0,802.0,838.0,804.0,838.0,836.0,132.0,834.0],"score":0.99,"text":"accurate real-time stereo matching methods as measured by the"},{"category_id":15,"poly":[134.0,830.0,619.0,830.0,619.0,862.0,134.0,862.0],"score":0.99,"text":"Middlebury stereo performance benchmark."},{"category_id":15,"poly":[887.0,1849.0,1568.0,1852.0,1568.0,1891.0,887.0,1888.0],"score":0.99,"text":" Disparity maps obtained using this simple strategy are often"},{"category_id":15,"poly":[862.0,1888.0,1568.0,1888.0,1568.0,1920.0,862.0,1920.0],"score":0.98,"text":"too noisy to be considered useable. To reduce the effects"},{"category_id":15,"poly":[864.0,1923.0,1568.0,1923.0,1568.0,1952.0,864.0,1952.0],"score":0.99,"text":"of noise and enforce spatial consistency of matches, local"},{"category_id":15,"poly":[862.0,1948.0,1568.0,1950.0,1568.0,1989.0,861.0,1987.0],"score":0.99,"text":"stereo algorithms consider arbitrarily shaped and sized support"},{"category_id":15,"poly":[864.0,1989.0,1568.0,1989.0,1568.0,2021.0,864.0,2021.0],"score":0.99,"text":"windows centered at each pixel of the reference image, and"}],"page_info":{"page_no":0,"height":2200,"width":1700}},{"layout_dets":[{"category_id":8,"poly":[962.3624267578125,1513.2073974609375,1465.4017333984375,1513.2073974609375,1465.4017333984375,1669.1397705078125,962.3624267578125,1669.1397705078125],"score":0.9999995231628418},{"category_id":9,"poly":[1530.72998046875,1101.879638671875,1565.2568359375,1101.879638671875,1565.2568359375,1130.8609619140625,1530.72998046875,1130.8609619140625],"score":0.9999992251396179},{"category_id":9,"poly":[1529.8787841796875,1575.843505859375,1565.931396484375,1575.843505859375,1565.931396484375,1607.2161865234375,1529.8787841796875,1607.2161865234375],"score":0.9999987483024597},{"category_id":1,"poly":[865.1971435546875,1684.040283203125,1566.561279296875,1684.040283203125,1566.561279296875,1813.7021484375,865.1971435546875,1813.7021484375],"score":0.9999987483024597},{"category_id":9,"poly":[1530.5263671875,1839.3990478515625,1565.1201171875,1839.3990478515625,1565.1201171875,1869.825439453125,1530.5263671875,1869.825439453125],"score":0.9999977946281433},{"category_id":8,"poly":[972.3255004882812,1075.85498046875,1461.2088623046875,1075.85498046875,1461.2088623046875,1155.465087890625,972.3255004882812,1155.465087890625],"score":0.999996542930603},{"category_id":1,"poly":[865.4874267578125,158.47100830078125,1565.84375,158.47100830078125,1565.84375,355.3230285644531,865.4874267578125,355.3230285644531],"score":0.9999960660934448},{"category_id":1,"poly":[133.51382446289062,158.21670532226562,835.5382080078125,158.21670532226562,835.5382080078125,558.8020629882812,133.51382446289062,558.8020629882812],"score":0.9999951124191284},{"category_id":1,"poly":[134.01239013671875,954.4151000976562,836.1470336914062,954.4151000976562,836.1470336914062,1618.77197265625,134.01239013671875,1618.77197265625],"score":0.9999947547912598},{"category_id":1,"poly":[134.4542999267578,558.8201904296875,834.2548828125,558.8201904296875,834.2548828125,954.7811279296875,134.4542999267578,954.7811279296875],"score":0.9999943971633911},{"category_id":1,"poly":[866.33642578125,421.84442138671875,1566.451904296875,421.84442138671875,1566.451904296875,787.1864624023438,866.33642578125,787.1864624023438],"score":0.9999930262565613},{"category_id":1,"poly":[864.974853515625,1167.92236328125,1567.0927734375,1167.92236328125,1567.0927734375,1298.29541015625,864.974853515625,1298.29541015625],"score":0.9999929666519165},{"category_id":1,"poly":[864.5220947265625,853.943359375,1565.82080078125,853.943359375,1565.82080078125,1080.8125,864.5220947265625,1080.8125],"score":0.9999923706054688},{"category_id":1,"poly":[865.4466552734375,1919.30615234375,1566.4720458984375,1919.30615234375,1566.4720458984375,2017.154541015625,865.4466552734375,2017.154541015625],"score":0.9999904036521912},{"category_id":1,"poly":[864.801513671875,1302.438232421875,1566.760986328125,1302.438232421875,1566.760986328125,1498.9681396484375,864.801513671875,1498.9681396484375],"score":0.9999889135360718},{"category_id":1,"poly":[133.34628295898438,1620.0596923828125,836.7553100585938,1620.0596923828125,836.7553100585938,2018.44873046875,133.34628295898438,2018.44873046875],"score":0.9999861717224121},{"category_id":0,"poly":[865.5296020507812,809.8997802734375,1302.7711181640625,809.8997802734375,1302.7711181640625,841.3140869140625,865.5296020507812,841.3140869140625],"score":0.9999798536300659},{"category_id":0,"poly":[1131.11181640625,378.66229248046875,1299.6181640625,378.66229248046875,1299.6181640625,409.04852294921875,1131.11181640625,409.04852294921875],"score":0.9999651908874512},{"category_id":8,"poly":[1003.5569458007812,1824.2362060546875,1420.7132568359375,1824.2362060546875,1420.7132568359375,1905.175048828125,1003.5569458007812,1905.175048828125],"score":0.999914288520813},{"category_id":14,"poly":[974,1076,1454,1076,1454,1155,974,1155],"score":0.94,"latex":"w(p,q)=\\exp{\\left(-\\frac{\\Delta_{g}(p,q)}{\\gamma_{g}}-\\frac{\\Delta_{c}(p,q)}{\\gamma_{c}}\\right)},"},{"category_id":14,"poly":[1006,1825,1423,1825,1423,1907,1006,1907],"score":0.94,"latex":"\\delta(q,\\bar{q})=\\sum_{c=\\{r,g,b\\}}\\operatorname*{min}(|q_{c}-\\bar{q}_{c}|,\\tau)."},{"category_id":14,"poly":[963,1510,1464,1510,1464,1671,963,1671],"score":0.93,"latex":"C(p,\\bar{p})=\\frac{\\displaystyle\\sum_{q\\in\\Omega_{p},\\bar{q}\\in\\Omega_{\\bar{p}}}w(p,q)w(\\bar{p},\\bar{q})\\delta(q,\\bar{q})}{\\displaystyle\\sum_{q\\in\\Omega_{p},\\bar{q}\\in\\Omega_{\\bar{p}}}w(p,q)w(\\bar{p},\\bar{q})}\\,,"},{"category_id":13,"poly":[1335,1166,1432,1166,1432,1200,1335,1200],"score":0.93,"latex":"\\Delta_{c}(p,q)"},{"category_id":13,"poly":[939,1166,1039,1166,1039,1201,939,1201],"score":0.93,"latex":"\\Delta_{g}(p,q)"},{"category_id":13,"poly":[1289,1683,1365,1683,1365,1717,1289,1717],"score":0.93,"latex":"\\delta(q,\\bar{q})"},{"category_id":13,"poly":[1362,1367,1441,1367,1441,1401,1362,1401],"score":0.92,"latex":"\\bar{p}\\in S_{p}"},{"category_id":13,"poly":[864,1019,951,1019,951,1053,864,1053],"score":0.92,"latex":"q\\in\\Omega_{p}"},{"category_id":13,"poly":[1351,953,1388,953,1388,987,1351,987],"score":0.9,"latex":"\\Omega_{p}"},{"category_id":13,"poly":[913,1467,949,1467,949,1501,913,1501],"score":0.89,"latex":"\\Omega_{\\bar{p}}"},{"category_id":13,"poly":[1531,1367,1565,1367,1565,1401,1531,1401],"score":0.89,"latex":"S_{p}"},{"category_id":13,"poly":[1528,1434,1565,1434,1565,1468,1528,1468],"score":0.89,"latex":"\\Omega_{p}"},{"category_id":13,"poly":[1485,1205,1516,1205,1516,1234,1485,1234],"score":0.88,"latex":"\\gamma_{g}"},{"category_id":13,"poly":[1159,1206,1178,1206,1178,1233,1159,1233],"score":0.82,"latex":"p"},{"category_id":13,"poly":[863,1238,893,1238,893,1266,863,1266],"score":0.82,"latex":"\\gamma_{c}"},{"category_id":13,"poly":[1177,1436,1196,1436,1196,1465,1177,1465],"score":0.8,"latex":"\\bar{p}"},{"category_id":13,"poly":[1371,1024,1391,1024,1391,1051,1371,1051],"score":0.8,"latex":"p"},{"category_id":13,"poly":[1540,1406,1558,1406,1558,1432,1540,1432],"score":0.8,"latex":"p"},{"category_id":13,"poly":[1447,1024,1465,1024,1465,1051,1447,1051],"score":0.79,"latex":"q"},{"category_id":13,"poly":[1101,1437,1121,1437,1121,1465,1101,1465],"score":0.79,"latex":"p"},{"category_id":13,"poly":[1389,1307,1407,1307,1407,1332,1389,1332],"score":0.79,"latex":"p"},{"category_id":13,"poly":[1230,1206,1247,1206,1247,1233,1230,1233],"score":0.78,"latex":"q"},{"category_id":13,"poly":[1029,1372,1048,1372,1048,1399,1029,1399],"score":0.78,"latex":"p"},{"category_id":13,"poly":[916,1752,934,1752,934,1782,916,1782],"score":0.76,"latex":"\\bar{q}"},{"category_id":13,"poly":[1407,1925,1425,1925,1425,1946,1407,1946],"score":0.75,"latex":"\\tau"},{"category_id":13,"poly":[1548,1722,1565,1722,1565,1749,1548,1749],"score":0.75,"latex":"q"},{"category_id":13,"poly":[1050,992,1068,992,1068,1018,1050,1018],"score":0.75,"latex":"p"},{"category_id":15,"poly":[864.0,1783.0,1298.0,1783.0,1298.0,1822.0,864.0,1822.0],"score":0.99,"text":"green, and blue components given by"},{"category_id":15,"poly":[866.0,1687.0,1288.0,1687.0,1288.0,1719.0,866.0,1719.0],"score":0.96,"text":"where the pixel dissimilarity metric"},{"category_id":15,"poly":[1366.0,1687.0,1564.0,1687.0,1564.0,1719.0,1366.0,1719.0],"score":0.97,"text":"ischosen as the"},{"category_id":15,"poly":[866.0,1751.0,915.0,1751.0,915.0,1783.0,866.0,1783.0],"score":1.0,"text":"and"},{"category_id":15,"poly":[935.0,1751.0,1564.0,1751.0,1564.0,1783.0,935.0,1783.0],"score":0.98,"text":". Here, the truncation of color difference for the red,"},{"category_id":15,"poly":[866.0,1719.0,1547.0,1719.0,1547.0,1749.0,866.0,1749.0],"score":0.99,"text":"sum of truncated absolute color differences between pixels"},{"category_id":15,"poly":[864.0,163.0,1568.0,163.0,1568.0,192.0,864.0,192.0],"score":1.0,"text":"temporal information, making it possible to process a temporal"},{"category_id":15,"poly":[859.0,188.0,1571.0,193.0,1570.0,229.0,859.0,225.0],"score":0.99,"text":" collection of cost volumes. The filtering operation was shown"},{"category_id":15,"poly":[864.0,229.0,1566.0,229.0,1566.0,261.0,864.0,261.0],"score":0.99,"text":"to preserve spatio-temporal edges present in the cost volumes,"},{"category_id":15,"poly":[859.0,261.0,1564.0,264.0,1564.0,296.0,859.0,293.0],"score":0.98,"text":" resulting in increased temporal consistency of disparity maps,"},{"category_id":15,"poly":[864.0,296.0,1566.0,296.0,1566.0,328.0,864.0,328.0],"score":0.99,"text":"greater robustness to image noise, and more accurate behavior"},{"category_id":15,"poly":[866.0,328.0,1160.0,328.0,1160.0,360.0,866.0,360.0],"score":1.0,"text":"around object boundaries."},{"category_id":15,"poly":[129.0,158.0,841.0,153.0,841.0,192.0,130.0,197.0],"score":0.99,"text":"aggregate cost values within the pixel neighborhoods defined"},{"category_id":15,"poly":[129.0,188.0,841.0,190.0,841.0,229.0,129.0,227.0],"score":0.99,"text":"by these windows. In 2005, Yoon and Kweon [4] proposed"},{"category_id":15,"poly":[132.0,229.0,838.0,229.0,838.0,261.0,132.0,261.0],"score":1.0,"text":"an adaptive matching cost aggregation scheme, which assigns"},{"category_id":15,"poly":[132.0,261.0,838.0,261.0,838.0,293.0,132.0,293.0],"score":0.98,"text":"a weight value to every pixel located in the support window"},{"category_id":15,"poly":[132.0,293.0,838.0,293.0,838.0,325.0,132.0,325.0],"score":0.98,"text":"of a given pixel of interest. The weight value is based on"},{"category_id":15,"poly":[132.0,328.0,836.0,328.0,836.0,360.0,132.0,360.0],"score":0.99,"text":"the spatial and color similarity between the pixel of interest"},{"category_id":15,"poly":[134.0,360.0,836.0,360.0,836.0,392.0,134.0,392.0],"score":1.0,"text":"and a pixel in its support window, and the aggregated cost is"},{"category_id":15,"poly":[134.0,394.0,836.0,394.0,836.0,426.0,134.0,426.0],"score":0.99,"text":"computed as a weighted average of the pixel-wise costs within"},{"category_id":15,"poly":[127.0,422.0,839.0,424.0,838.0,463.0,127.0,461.0],"score":0.98,"text":" the considered support window. The edge-preserving nature"},{"category_id":15,"poly":[129.0,456.0,838.0,454.0,838.0,493.0,129.0,495.0],"score":0.99,"text":" and matching accuracy of adaptive support weights have made"},{"category_id":15,"poly":[132.0,490.0,841.0,490.0,841.0,529.0,132.0,529.0],"score":0.99,"text":"them one of the most popular choices for cost aggregation in"},{"category_id":15,"poly":[132.0,527.0,797.0,527.0,797.0,559.0,132.0,559.0],"score":0.97,"text":"recently proposed stereo matching algorithms [3], [5]-[8]."},{"category_id":15,"poly":[157.0,958.0,836.0,958.0,836.0,988.0,157.0,988.0],"score":0.99,"text":"It has been demonstrated that the performance of stereo"},{"category_id":15,"poly":[132.0,990.0,838.0,990.0,838.0,1022.0,132.0,1022.0],"score":0.99,"text":"algorithms designed to match a single pair of images can"},{"category_id":15,"poly":[132.0,1024.0,836.0,1024.0,836.0,1056.0,132.0,1056.0],"score":0.99,"text":"be adapted to take advantage of the temporal dependencies"},{"category_id":15,"poly":[129.0,1054.0,838.0,1054.0,838.0,1093.0,129.0,1093.0],"score":0.97,"text":"available in stereo video sequences. Early proposed solutions"},{"category_id":15,"poly":[132.0,1091.0,836.0,1091.0,836.0,1123.0,132.0,1123.0],"score":0.99,"text":"to temporal stereo matching attempted to average matching"},{"category_id":15,"poly":[134.0,1123.0,836.0,1123.0,836.0,1155.0,134.0,1155.0],"score":0.99,"text":"costs across subsequent frames of a video sequence [13],"},{"category_id":15,"poly":[129.0,1153.0,841.0,1150.0,841.0,1189.0,129.0,1192.0],"score":0.98,"text":"[14]. Attempts have been made to integrate estimation of"},{"category_id":15,"poly":[134.0,1192.0,838.0,1192.0,838.0,1224.0,134.0,1224.0],"score":0.99,"text":"motion fields (optical flow) into temporal stereo matching. The"},{"category_id":15,"poly":[132.0,1224.0,838.0,1224.0,838.0,1256.0,132.0,1256.0],"score":0.99,"text":"methods of [15] and [16] perform smoothing of disparities"},{"category_id":15,"poly":[129.0,1254.0,841.0,1254.0,841.0,1292.0,129.0,1292.0],"score":0.99,"text":" along motion vectors recovered from the video sequence. The"},{"category_id":15,"poly":[132.0,1290.0,838.0,1290.0,838.0,1322.0,132.0,1322.0],"score":0.99,"text":"estimation of the motion field, however, prevents real-time"},{"category_id":15,"poly":[132.0,1325.0,838.0,1325.0,838.0,1354.0,132.0,1354.0],"score":0.99,"text":"implementation, since state-of-the-art optical flow algorithms"},{"category_id":15,"poly":[129.0,1354.0,841.0,1354.0,841.0,1393.0,129.0,1393.0],"score":0.99,"text":" do not, in general, approach real-time frame rates. In a related"},{"category_id":15,"poly":[129.0,1386.0,841.0,1384.0,841.0,1423.0,129.0,1425.0],"score":0.99,"text":"approach, Sizintsev and Wildes [17], [18] used steerable"},{"category_id":15,"poly":[134.0,1423.0,836.0,1423.0,836.0,1455.0,134.0,1455.0],"score":0.99,"text":"filters to obtain descriptors characterizing motion of image"},{"category_id":15,"poly":[134.0,1455.0,836.0,1455.0,836.0,1487.0,134.0,1487.0],"score":0.99,"text":"features in both space and time. Unlike traditional algorithms,"},{"category_id":15,"poly":[132.0,1490.0,838.0,1490.0,838.0,1522.0,132.0,1522.0],"score":0.98,"text":"their method performs matching on spatio-temporal motion"},{"category_id":15,"poly":[129.0,1519.0,841.0,1517.0,841.0,1556.0,129.0,1558.0],"score":0.99,"text":" descriptors, rather than on pure pixel intensity values, which"},{"category_id":15,"poly":[132.0,1554.0,841.0,1554.0,841.0,1593.0,132.0,1593.0],"score":0.99,"text":"leads to improved temporal coherence of disparity maps at the"},{"category_id":15,"poly":[132.0,1586.0,698.0,1586.0,698.0,1618.0,132.0,1618.0],"score":0.99,"text":"cost of reduced accuracy at depth discontinuities."},{"category_id":15,"poly":[159.0,559.0,838.0,559.0,838.0,591.0,159.0,591.0],"score":0.99,"text":"Recently, Rheman et al. [9], [10] have revisited the cost"},{"category_id":15,"poly":[132.0,594.0,838.0,589.0,839.0,621.0,132.0,626.0],"score":1.0,"text":"aggregation step of stereo algorithms, and demonstrated that"},{"category_id":15,"poly":[132.0,626.0,838.0,626.0,838.0,658.0,132.0,658.0],"score":0.99,"text":"cost aggregation can be performed by filtering of subsequent"},{"category_id":15,"poly":[134.0,660.0,834.0,660.0,834.0,692.0,134.0,692.0],"score":1.0,"text":"layers of the initially computed matching cost volume. In par-"},{"category_id":15,"poly":[132.0,692.0,836.0,692.0,836.0,724.0,132.0,724.0],"score":0.99,"text":"ticular, the edge-aware image filters, such as the bilateral filter"},{"category_id":15,"poly":[127.0,719.0,839.0,724.0,838.0,761.0,127.0,756.0],"score":0.99,"text":" of Tomasi and Manducci [11] or the guided filter of He [12],"},{"category_id":15,"poly":[132.0,759.0,838.0,759.0,838.0,791.0,132.0,791.0],"score":0.98,"text":"have been rendered useful for the problem of matching cost"},{"category_id":15,"poly":[132.0,793.0,838.0,791.0,838.0,823.0,132.0,825.0],"score":0.99,"text":"aggregation, enabling stereo algorithms to correctly recover"},{"category_id":15,"poly":[134.0,825.0,838.0,825.0,838.0,857.0,134.0,857.0],"score":0.98,"text":"disparities along object boundaries. In fact, Yoon and Kweon's"},{"category_id":15,"poly":[134.0,859.0,838.0,859.0,838.0,891.0,134.0,891.0],"score":1.0,"text":"adaptive support-weight cost aggregation scheme is equivalent"},{"category_id":15,"poly":[132.0,891.0,838.0,891.0,838.0,924.0,132.0,924.0],"score":0.98,"text":"to the application of the so-called joint bilateral filter to the"},{"category_id":15,"poly":[134.0,924.0,547.0,924.0,547.0,956.0,134.0,956.0],"score":1.0,"text":"layers of the matching cost volume."},{"category_id":15,"poly":[889.0,422.0,1568.0,424.0,1568.0,456.0,889.0,454.0],"score":0.98,"text":"The proposed temporal stereo matching algorithm is an"},{"category_id":15,"poly":[862.0,456.0,1571.0,456.0,1571.0,495.0,862.0,495.0],"score":1.0,"text":"extension of the real-time iterative adaptive support-weight"},{"category_id":15,"poly":[864.0,490.0,1568.0,490.0,1568.0,522.0,864.0,522.0],"score":0.99,"text":"algorithm described in [3]. In addition to real-time two-"},{"category_id":15,"poly":[864.0,525.0,1566.0,525.0,1566.0,557.0,864.0,557.0],"score":1.0,"text":"pass aggregation of the cost values in the spatial domain,"},{"category_id":15,"poly":[864.0,557.0,1568.0,557.0,1568.0,589.0,864.0,589.0],"score":0.99,"text":"the proposed algorithm enhances stereo matching on video"},{"category_id":15,"poly":[866.0,594.0,1566.0,594.0,1566.0,626.0,866.0,626.0],"score":0.97,"text":"sequences by aggregating costs along the time dimension."},{"category_id":15,"poly":[864.0,626.0,1568.0,626.0,1568.0,658.0,864.0,658.0],"score":1.0,"text":"The operation of the algorithm has been divided into four"},{"category_id":15,"poly":[866.0,660.0,1568.0,660.0,1568.0,692.0,866.0,692.0],"score":0.99,"text":"stages: 1) two-pass spatial cost aggregation, 2) temporal cost"},{"category_id":15,"poly":[862.0,688.0,1568.0,685.0,1568.0,724.0,862.0,727.0],"score":1.0,"text":"aggregation, 3) disparity selection and confidence assessment,"},{"category_id":15,"poly":[866.0,724.0,1568.0,724.0,1568.0,756.0,866.0,756.0],"score":1.0,"text":"and 4) iterative disparity refinement. In the following, each of"},{"category_id":15,"poly":[864.0,759.0,1254.0,759.0,1254.0,791.0,864.0,791.0],"score":1.0,"text":"these stages is described in detail."},{"category_id":15,"poly":[860.0,1265.0,1194.0,1270.0,1194.0,1306.0,859.0,1301.0],"score":0.99,"text":" color similarity, respectively."},{"category_id":15,"poly":[1433.0,1169.0,1566.0,1169.0,1566.0,1201.0,1433.0,1201.0],"score":0.98,"text":"is the color"},{"category_id":15,"poly":[864.0,1169.0,938.0,1169.0,938.0,1201.0,864.0,1201.0],"score":1.0,"text":"where"},{"category_id":15,"poly":[1040.0,1169.0,1334.0,1169.0,1334.0,1201.0,1040.0,1201.0],"score":0.98,"text":"is the geometric distance,"},{"category_id":15,"poly":[1517.0,1196.0,1566.0,1201.0,1566.0,1240.0,1517.0,1235.0],"score":1.0,"text":"and"},{"category_id":15,"poly":[862.0,1196.0,1158.0,1201.0,1158.0,1240.0,861.0,1235.0],"score":1.0,"text":"difference between pixels"},{"category_id":15,"poly":[894.0,1233.0,1566.0,1231.0,1566.0,1270.0,894.0,1272.0],"score":0.97,"text":"regulate the strength of grouping by geometric distance and"},{"category_id":15,"poly":[1179.0,1196.0,1229.0,1201.0,1229.0,1240.0,1179.0,1235.0],"score":1.0,"text":"and"},{"category_id":15,"poly":[1248.0,1196.0,1484.0,1201.0,1484.0,1240.0,1248.0,1235.0],"score":0.99,"text":", and the coefficients"},{"category_id":15,"poly":[887.0,848.0,1568.0,850.0,1568.0,889.0,887.0,887.0],"score":0.99,"text":" Humans group shapes by observing the geometric distance"},{"category_id":15,"poly":[859.0,885.0,1568.0,882.0,1568.0,921.0,859.0,924.0],"score":0.98,"text":" and color similarity of points in space. To mimic this vi-"},{"category_id":15,"poly":[864.0,921.0,1568.0,921.0,1568.0,953.0,864.0,953.0],"score":0.99,"text":"sual grouping, the adaptive support-weight stereo matching"},{"category_id":15,"poly":[864.0,1054.0,899.0,1054.0,899.0,1084.0,864.0,1084.0],"score":1.0,"text":"by"},{"category_id":15,"poly":[866.0,956.0,1350.0,956.0,1350.0,988.0,866.0,988.0],"score":0.98,"text":"algorithm [4] considers a support window"},{"category_id":15,"poly":[1389.0,956.0,1566.0,956.0,1566.0,988.0,1389.0,988.0],"score":0.98,"text":" centered at the"},{"category_id":15,"poly":[952.0,1022.0,1370.0,1022.0,1370.0,1054.0,952.0,1054.0],"score":0.98,"text":". The support weight relating pixels"},{"category_id":15,"poly":[1392.0,1022.0,1446.0,1022.0,1446.0,1054.0,1392.0,1054.0],"score":1.0,"text":"and"},{"category_id":15,"poly":[1466.0,1022.0,1566.0,1022.0,1566.0,1054.0,1466.0,1054.0],"score":0.98,"text":"is given"},{"category_id":15,"poly":[866.0,990.0,1049.0,990.0,1049.0,1022.0,866.0,1022.0],"score":1.0,"text":"pixel of interest"},{"category_id":15,"poly":[1069.0,990.0,1566.0,990.0,1566.0,1022.0,1069.0,1022.0],"score":1.0,"text":", and assigns a support weight to each pixel"},{"category_id":15,"poly":[862.0,1948.0,1568.0,1950.0,1568.0,1989.0,861.0,1987.0],"score":0.98,"text":"vides additional robustness to outliers. Rather than evaluating"},{"category_id":15,"poly":[864.0,1989.0,1566.0,1989.0,1566.0,2021.0,864.0,2021.0],"score":0.98,"text":"Equation (2) directly, real-time algorithms often approximate"},{"category_id":15,"poly":[862.0,1920.0,1406.0,1920.0,1406.0,1952.0,862.0,1952.0],"score":0.99,"text":"This limits each of their magnitudes to at most"},{"category_id":15,"poly":[1426.0,1920.0,1561.0,1920.0,1561.0,1952.0,1426.0,1952.0],"score":0.96,"text":",whichpro-"},{"category_id":15,"poly":[859.0,1331.0,1571.0,1334.0,1571.0,1373.0,859.0,1370.0],"score":0.98,"text":" iterative adaptive support-weight algorithm evaluates matching"},{"category_id":15,"poly":[859.0,1464.0,912.0,1467.0,912.0,1506.0,859.0,1503.0],"score":1.0,"text":"and"},{"category_id":15,"poly":[950.0,1464.0,1474.0,1467.0,1474.0,1506.0,950.0,1503.0],"score":1.0,"text":", the initial matching cost is aggregated using"},{"category_id":15,"poly":[1442.0,1370.0,1530.0,1370.0,1530.0,1402.0,1442.0,1402.0],"score":0.98,"text":", where"},{"category_id":15,"poly":[1197.0,1437.0,1527.0,1437.0,1527.0,1469.0,1197.0,1469.0],"score":0.97,"text":", and their support windows"},{"category_id":15,"poly":[866.0,1402.0,1539.0,1402.0,1539.0,1435.0,866.0,1435.0],"score":1.0,"text":"denotes a set of matching candidates associated with pixel"},{"category_id":15,"poly":[864.0,1437.0,1100.0,1437.0,1100.0,1469.0,864.0,1469.0],"score":0.97,"text":"For a pair of pixels"},{"category_id":15,"poly":[1122.0,1437.0,1176.0,1437.0,1176.0,1469.0,1122.0,1469.0],"score":0.94,"text":" and"},{"category_id":15,"poly":[887.0,1299.0,1388.0,1304.0,1388.0,1336.0,887.0,1331.0],"score":0.96,"text":" To identify a match for the pixel of interest"},{"category_id":15,"poly":[1408.0,1299.0,1568.0,1304.0,1568.0,1336.0,1408.0,1331.0],"score":1.0,"text":", the real-time"},{"category_id":15,"poly":[864.0,1370.0,1028.0,1370.0,1028.0,1402.0,864.0,1402.0],"score":1.0,"text":"costs between"},{"category_id":15,"poly":[1049.0,1370.0,1361.0,1370.0,1361.0,1402.0,1049.0,1402.0],"score":0.99,"text":" and every match candidate"},{"category_id":15,"poly":[160.0,1618.0,836.0,1623.0,836.0,1655.0,159.0,1650.0],"score":0.99,"text":"Most recently, local stereo algorithms based on edge-aware"},{"category_id":15,"poly":[127.0,1650.0,841.0,1652.0,841.0,1691.0,127.0,1689.0],"score":0.97,"text":" filters were extended to incorporate temporal evidence into"},{"category_id":15,"poly":[132.0,1687.0,836.0,1687.0,836.0,1719.0,132.0,1719.0],"score":0.97,"text":"the matching process. The method of Richardt et al. [19]"},{"category_id":15,"poly":[134.0,1723.0,838.0,1723.0,838.0,1753.0,134.0,1753.0],"score":0.99,"text":"employs a variant of the bilateral grid [20] implemented on"},{"category_id":15,"poly":[134.0,1755.0,838.0,1755.0,838.0,1788.0,134.0,1788.0],"score":0.99,"text":"graphics hardware, which accelerates cost aggregation and"},{"category_id":15,"poly":[134.0,1788.0,838.0,1788.0,838.0,1820.0,134.0,1820.0],"score":1.0,"text":"allows for weighted propagation of pixel dissimilarity metrics"},{"category_id":15,"poly":[132.0,1822.0,838.0,1822.0,838.0,1854.0,132.0,1854.0],"score":0.99,"text":"from previous frames to the current one. Although this method"},{"category_id":15,"poly":[129.0,1856.0,838.0,1856.0,838.0,1888.0,129.0,1888.0],"score":1.0,"text":" outperforms the baseline frame-to-frame approach, the amount"},{"category_id":15,"poly":[132.0,1888.0,838.0,1888.0,838.0,1920.0,132.0,1920.0],"score":0.97,"text":"of hardware memory necessary to construct the bilateral grid"},{"category_id":15,"poly":[127.0,1916.0,841.0,1918.0,841.0,1957.0,127.0,1955.0],"score":0.99,"text":"limits its application to single-channel, i.e., grayscale images "},{"category_id":15,"poly":[132.0,1955.0,838.0,1955.0,838.0,1985.0,132.0,1985.0],"score":0.99,"text":"only. Hosni et al. [10], on the other hand, reformulated kernels"},{"category_id":15,"poly":[132.0,1989.0,838.0,1989.0,838.0,2021.0,132.0,2021.0],"score":0.99,"text":"of the guided image filter to operate on both spatial and"},{"category_id":15,"poly":[859.0,809.0,1307.0,809.0,1307.0,848.0,859.0,848.0],"score":0.99,"text":"A. Two-Pass Spatial Cost Aggregation"},{"category_id":15,"poly":[1129.0,376.0,1300.0,376.0,1300.0,417.0,1129.0,417.0],"score":0.94,"text":"III. METHOD"}],"page_info":{"page_no":1,"height":2200,"width":1700}},{"layout_dets":[{"category_id":1,"poly":[865.5088500976562,856.5537109375,1567.692626953125,856.5537109375,1567.692626953125,1420.9698486328125,865.5088500976562,1420.9698486328125],"score":0.9999963045120239},{"category_id":8,"poly":[281.1294860839844,1001.0513916015625,689.37451171875,1001.0513916015625,689.37451171875,1075.8765869140625,281.1294860839844,1075.8765869140625],"score":0.9999961256980896},{"category_id":1,"poly":[133.53353881835938,158.6427459716797,836.7297973632812,158.6427459716797,836.7297973632812,390.48828125,133.53353881835938,390.48828125],"score":0.9999960660934448},{"category_id":8,"poly":[145.77777099609375,1839.6416015625,803.4192504882812,1839.6416015625,803.4192504882812,1993.239013671875,145.77777099609375,1993.239013671875],"score":0.9999958872795105},{"category_id":1,"poly":[864.9884643554688,1420.8831787109375,1567.3118896484375,1420.8831787109375,1567.3118896484375,2023.257080078125,864.9884643554688,2023.257080078125],"score":0.9999951124191284},{"category_id":9,"poly":[1529.267333984375,388.6717834472656,1565.1744384765625,388.6717834472656,1565.1744384765625,416.4899597167969,1529.267333984375,416.4899597167969],"score":0.9999918937683105},{"category_id":9,"poly":[800.3933715820312,1551.524169921875,833.2618408203125,1551.524169921875,833.2618408203125,1582.073486328125,800.3933715820312,1582.073486328125],"score":0.9999911189079285},{"category_id":1,"poly":[864.3720092773438,200.97483825683594,1565.6871337890625,200.97483825683594,1565.6871337890625,365.6230163574219,864.3720092773438,365.6230163574219],"score":0.9999903440475464},{"category_id":1,"poly":[134.87628173828125,1369.5762939453125,835.0336303710938,1369.5762939453125,835.0336303710938,1533.884765625,134.87628173828125,1533.884765625],"score":0.9999880790710449},{"category_id":1,"poly":[134.59988403320312,444.5299377441406,836.5606079101562,444.5299377441406,836.5606079101562,709.0791015625,134.59988403320312,709.0791015625],"score":0.999987006187439},{"category_id":1,"poly":[134.15472412109375,1084.4288330078125,836.2360229492188,1084.4288330078125,836.2360229492188,1314.6600341796875,134.15472412109375,1314.6600341796875],"score":0.9999866485595703},{"category_id":9,"poly":[800.6007690429688,1023.1047973632812,833.2154541015625,1023.1047973632812,833.2154541015625,1055.7227783203125,800.6007690429688,1055.7227783203125],"score":0.9999839663505554},{"category_id":8,"poly":[948.4016723632812,372.03607177734375,1486.11279296875,372.03607177734375,1486.11279296875,449.3696594238281,948.4016723632812,449.3696594238281],"score":0.9999831914901733},{"category_id":8,"poly":[145.31065368652344,714.4036254882812,820.3599853515625,714.4036254882812,820.3599853515625,791.855712890625,145.31065368652344,791.855712890625],"score":0.9999772906303406},{"category_id":1,"poly":[863.8760986328125,599.6033325195312,1566.84619140625,599.6033325195312,1566.84619140625,797.44189453125,863.8760986328125,797.44189453125],"score":0.999976396560669},{"category_id":1,"poly":[864.925537109375,464.9669189453125,1565.212158203125,464.9669189453125,1565.212158203125,529.045654296875,864.925537109375,529.045654296875],"score":0.999973475933075},{"category_id":1,"poly":[133.88735961914062,797.7457885742188,835.5986328125,797.7457885742188,835.5986328125,994.4456176757812,133.88735961914062,994.4456176757812],"score":0.9999661445617676},{"category_id":1,"poly":[134.8787841796875,1615.116455078125,835.4554443359375,1615.116455078125,835.4554443359375,1815.4564208984375,134.8787841796875,1815.4564208984375],"score":0.9999580383300781},{"category_id":9,"poly":[1530.1783447265625,550.1576538085938,1564.607177734375,550.1576538085938,1564.607177734375,578.6950073242188,1530.1783447265625,578.6950073242188],"score":0.9999532103538513},{"category_id":9,"poly":[801.0740966796875,738.4259643554688,834.7449340820312,738.4259643554688,834.7449340820312,770.4969482421875,801.0740966796875,770.4969482421875],"score":0.9996598958969116},{"category_id":0,"poly":[1134.302490234375,815.6021728515625,1295.3885498046875,815.6021728515625,1295.3885498046875,844.6544799804688,1134.302490234375,844.6544799804688],"score":0.9994980096817017},{"category_id":9,"poly":[798.6090698242188,1986.7332763671875,834.5460205078125,1986.7332763671875,834.5460205078125,2017.6595458984375,798.6090698242188,2017.6595458984375],"score":0.9992558360099792},{"category_id":0,"poly":[135.0093994140625,406.12335205078125,475.6328125,406.12335205078125,475.6328125,437.4545593261719,135.0093994140625,437.4545593261719],"score":0.9990860819816589},{"category_id":8,"poly":[1029.3924560546875,541.857177734375,1400.174072265625,541.857177734375,1400.174072265625,585.1640625,1029.3924560546875,585.1640625],"score":0.9979717135429382},{"category_id":0,"poly":[133.26077270507812,1330.139892578125,713.5426635742188,1330.139892578125,713.5426635742188,1363.1341552734375,133.26077270507812,1363.1341552734375],"score":0.9967154860496521},{"category_id":8,"poly":[338.6681823730469,1547.7218017578125,626.6519775390625,1547.7218017578125,626.6519775390625,1604.587646484375,338.6681823730469,1604.587646484375],"score":0.9945433139801025},{"category_id":1,"poly":[864.5469970703125,160.16702270507812,1251.313720703125,160.16702270507812,1251.313720703125,190.15760803222656,864.5469970703125,190.15760803222656],"score":0.9902143478393555},{"category_id":13,"poly":[550,577,648,577,648,612,550,612],"score":0.95,"latex":"C_{a}(p,\\bar{p})"},{"category_id":13,"poly":[183,1780,304,1780,304,1813,183,1813],"score":0.95,"latex":"p^{\\prime}=m(\\bar{p})"},{"category_id":14,"poly":[279,1000,687,1000,687,1078,279,1078],"score":0.95,"latex":"w_{t}(p,p_{t-1})=\\exp\\bigg({-\\frac{\\Delta_{c}(p,p_{t-1})}{\\gamma_{t}}}\\bigg),"},{"category_id":14,"poly":[147,1843,820,1843,820,1992,147,1992],"score":0.94,"latex":"F_{p}=\\left\\{\\begin{array}{l l}{\\underset{\\bar{p}\\in S_{p}\\setminus m(p)}{\\mathrm{min}}\\,C(p,\\bar{p})-\\underset{\\bar{p}\\in S_{p}}{\\mathrm{min}}\\,C(p,\\bar{p})}\\\\ {\\underset{\\bar{p}\\in S_{p}\\setminus m(p)}{\\mathrm{min}}\\,C(p,\\bar{p})}&{|d_{p}-d_{p^{\\prime}}|\\leq1}\\\\ {0,}&{\\mathrm{otherwise}}\\end{array}\\right.."},{"category_id":14,"poly":[340,1546,628,1546,628,1608,340,1608],"score":0.93,"latex":"m(p)=\\underset{\\bar{p}\\in S_{p}}{\\mathrm{argmin}}\\,C(p,\\bar{p})\\,."},{"category_id":13,"poly":[321,830,443,830,443,864,321,864],"score":0.93,"latex":"w_{t}(p,p_{t-1})"},{"category_id":13,"poly":[581,1713,694,1713,694,1747,581,1747],"score":0.93,"latex":"{\\bar{p}}=m(p)"},{"category_id":14,"poly":[947,373,1478,373,1478,454,947,454],"score":0.93,"latex":"\\Lambda^{i}(p,\\bar{p})=\\alpha\\times\\sum_{q\\in\\Omega_{p}}w(p,q)F_{q}^{i-1}\\left|D_{q}^{i-1}-d_{p}\\right|\\,,"},{"category_id":13,"poly":[426,445,512,445,512,479,426,479],"score":0.93,"latex":"C(p,{\\bar{p}})"},{"category_id":13,"poly":[337,356,414,356,414,391,337,391],"score":0.93,"latex":"\\mathcal{O}(\\omega^{2})"},{"category_id":13,"poly":[1341,730,1565,730,1565,765,1341,765],"score":0.92,"latex":"C_{a}(p,\\bar{p})\\gets C(p,\\bar{p})"},{"category_id":13,"poly":[629,1436,691,1436,691,1470,629,1470],"score":0.92,"latex":"m(p)"},{"category_id":13,"poly":[277,1469,361,1469,361,1504,277,1504],"score":0.92,"latex":"\\bar{p}\\in S_{p}"},{"category_id":14,"poly":[1030,541,1398,541,1398,582,1030,582],"score":0.92,"latex":"C^{i}(p,\\bar{p})=C^{0}(p,\\bar{p})+{\\Lambda^{i}}(p,\\bar{p})\\,,"},{"category_id":13,"poly":[453,356,518,356,518,391,453,391],"score":0.91,"latex":"\\mathcal{O}(\\omega)"},{"category_id":14,"poly":[146,714,787,714,787,791,146,791],"score":0.91,"latex":"C(p,\\bar{p})\\gets\\frac{(1-\\lambda)\\cdot C(p,\\bar{p})+\\lambda\\cdot w_{t}(p,p_{t-1})\\cdot C_{a}(p,\\bar{p})}{(1-\\lambda)+\\lambda\\cdot w_{t}(p,p_{t-1})},"},{"category_id":13,"poly":[1095,231,1134,231,1134,270,1095,270],"score":0.9,"latex":"D_{p}^{i}"},{"category_id":13,"poly":[1313,1752,1447,1752,1447,1783,1313,1783],"score":0.89,"latex":"640~\\times~480"},{"category_id":13,"poly":[593,1782,627,1782,627,1815,593,1815],"score":0.89,"latex":"F_{p}"},{"category_id":13,"poly":[133,326,209,326,209,355,133,355],"score":0.88,"latex":"\\omega\\times\\omega"},{"category_id":13,"poly":[208,1089,236,1089,236,1116,208,1116],"score":0.85,"latex":"\\gamma_{t}"},{"category_id":13,"poly":[1466,769,1484,769,1484,797,1466,797],"score":0.83,"latex":"\\bar{p}"},{"category_id":13,"poly":[133,935,177,935,177,963,133,963],"score":0.83,"latex":"p_{t-1}"},{"category_id":13,"poly":[608,1753,627,1753,627,1779,608,1779],"score":0.81,"latex":"p"},{"category_id":13,"poly":[491,799,511,799,511,825,491,825],"score":0.81,"latex":"\\lambda"},{"category_id":13,"poly":[1018,770,1037,770,1037,796,1018,796],"score":0.81,"latex":"p"},{"category_id":13,"poly":[1086,470,1107,470,1107,491,1086,491],"score":0.8,"latex":"\\alpha"},{"category_id":13,"poly":[466,901,485,901,485,929,466,929],"score":0.8,"latex":"p"},{"category_id":13,"poly":[208,484,227,484,227,511,208,511],"score":0.79,"latex":"p"},{"category_id":13,"poly":[462,1443,480,1443,480,1468,462,1468],"score":0.77,"latex":"p"},{"category_id":13,"poly":[266,514,288,514,288,544,266,544],"score":0.77,"latex":"\\bar{p}"},{"category_id":13,"poly":[816,1716,836,1716,836,1746,816,1746],"score":0.73,"latex":"\\bar{p}"},{"category_id":13,"poly":[132,405,154,405,154,432,132,432],"score":0.27,"latex":"B"},{"category_id":13,"poly":[862,160,887,160,887,187,862,187],"score":0.26,"latex":"D"},{"category_id":15,"poly":[887.0,852.0,1568.0,855.0,1568.0,894.0,887.0,891.0],"score":0.98,"text":" The speed and accuracy of real-time stereo matching al-"},{"category_id":15,"poly":[864.0,891.0,1566.0,891.0,1566.0,924.0,864.0,924.0],"score":0.99,"text":"gorithms are traditionally demonstrated using still-frame im-"},{"category_id":15,"poly":[859.0,921.0,1571.0,919.0,1571.0,958.0,859.0,960.0],"score":0.97,"text":" ages from the Middlebury stereo benchmark [1], [2]. Still"},{"category_id":15,"poly":[862.0,956.0,1568.0,958.0,1568.0,990.0,862.0,988.0],"score":0.99,"text":"frames, however, are insufficient for evaluating stereo match-"},{"category_id":15,"poly":[864.0,992.0,1571.0,992.0,1571.0,1024.0,864.0,1024.0],"score":1.0,"text":"ing algorithms that incorporate frame-to-frame prediction to"},{"category_id":15,"poly":[864.0,1027.0,1568.0,1027.0,1568.0,1059.0,864.0,1059.0],"score":0.97,"text":"enhance matching accuracy. An alternative approach is to"},{"category_id":15,"poly":[864.0,1059.0,1566.0,1059.0,1566.0,1089.0,864.0,1089.0],"score":0.99,"text":"use a stereo video sequence with a ground truth disparity"},{"category_id":15,"poly":[862.0,1091.0,1566.0,1091.0,1566.0,1123.0,862.0,1123.0],"score":1.0,"text":"for each frame. Obtaining the ground truth disparity of real"},{"category_id":15,"poly":[866.0,1125.0,1566.0,1125.0,1566.0,1157.0,866.0,1157.0],"score":0.98,"text":"world video sequences is a difficult undertaking due to the"},{"category_id":15,"poly":[859.0,1153.0,1568.0,1155.0,1568.0,1194.0,859.0,1192.0],"score":0.99,"text":"high frame rate of video and limitations in depth sensing-"},{"category_id":15,"poly":[864.0,1192.0,1568.0,1192.0,1568.0,1224.0,864.0,1224.0],"score":0.99,"text":"technology. To address the need for stereo video with ground"},{"category_id":15,"poly":[864.0,1224.0,1568.0,1224.0,1568.0,1256.0,864.0,1256.0],"score":0.99,"text":"truth disparities, five pairs of synthetic stereo video sequences"},{"category_id":15,"poly":[864.0,1258.0,1568.0,1258.0,1568.0,1290.0,864.0,1290.0],"score":0.99,"text":"of a computer-generated scene were given in [19]. While these"},{"category_id":15,"poly":[864.0,1290.0,1566.0,1290.0,1566.0,1322.0,864.0,1322.0],"score":1.0,"text":"videos incorporate a sufficient amount of movement variation,"},{"category_id":15,"poly":[862.0,1325.0,1568.0,1325.0,1568.0,1357.0,862.0,1357.0],"score":0.99,"text":"they were generated from relatively simple models using low-"},{"category_id":15,"poly":[862.0,1359.0,1571.0,1359.0,1571.0,1389.0,862.0,1389.0],"score":0.99,"text":"resolution rendering, and they do not provide occlusion or"},{"category_id":15,"poly":[862.0,1386.0,1088.0,1394.0,1087.0,1426.0,861.0,1418.0],"score":0.98,"text":"discontinuity maps."},{"category_id":15,"poly":[129.0,156.0,839.0,158.0,838.0,197.0,129.0,195.0],"score":0.99,"text":"the matching cost by performing two-pass aggregation using"},{"category_id":15,"poly":[130.0,188.0,841.0,193.0,841.0,229.0,129.0,225.0],"score":0.98,"text":"two orthogonal 1D windows [5], [6], [8]. The two-pass method "},{"category_id":15,"poly":[129.0,225.0,841.0,222.0,841.0,261.0,129.0,264.0],"score":0.99,"text":"first aggregates matching costs in the vertical direction, and"},{"category_id":15,"poly":[134.0,261.0,838.0,261.0,838.0,293.0,134.0,293.0],"score":0.99,"text":"then computes a weighted sum of the aggregated costs in the"},{"category_id":15,"poly":[132.0,291.0,838.0,291.0,838.0,330.0,132.0,330.0],"score":0.99,"text":"horizontal direction. Given that support regions are of size"},{"category_id":15,"poly":[136.0,360.0,336.0,360.0,336.0,392.0,136.0,392.0],"score":0.99,"text":"aggregation from"},{"category_id":15,"poly":[415.0,360.0,452.0,360.0,452.0,392.0,415.0,392.0],"score":0.98,"text":"to"},{"category_id":15,"poly":[210.0,321.0,836.0,321.0,836.0,360.0,210.0,360.0],"score":0.98,"text":", the two-pass method reduces the complexity of cost"},{"category_id":15,"poly":[887.0,1416.0,1571.0,1419.0,1571.0,1458.0,887.0,1455.0],"score":0.98,"text":" To evaluate the performance of temporal aggregation, a"},{"category_id":15,"poly":[862.0,1453.0,1566.0,1453.0,1566.0,1485.0,862.0,1485.0],"score":0.98,"text":"new synthetic stereo video sequence is introduced along with"},{"category_id":15,"poly":[862.0,1490.0,1566.0,1487.0,1566.0,1519.0,862.0,1522.0],"score":0.99,"text":"corresponding disparity maps, occlusion maps, and disconti-"},{"category_id":15,"poly":[862.0,1519.0,1571.0,1519.0,1571.0,1558.0,862.0,1558.0],"score":0.99,"text":"nuity maps for evaluating the performance of temporal stereo"},{"category_id":15,"poly":[864.0,1556.0,1568.0,1556.0,1568.0,1588.0,864.0,1588.0],"score":1.0,"text":"matching algorithms. To create the video sequence, a complex"},{"category_id":15,"poly":[864.0,1590.0,1568.0,1590.0,1568.0,1620.0,864.0,1620.0],"score":0.99,"text":"scene was constructed using Google Sketchup and a pair"},{"category_id":15,"poly":[864.0,1622.0,1568.0,1622.0,1568.0,1655.0,864.0,1655.0],"score":0.99,"text":"of animated paths were rendered photorealistically using the"},{"category_id":15,"poly":[859.0,1650.0,1571.0,1652.0,1571.0,1691.0,859.0,1689.0],"score":0.99,"text":" Kerkythea rendering software. Realistic material properties"},{"category_id":15,"poly":[864.0,1689.0,1566.0,1689.0,1566.0,1721.0,864.0,1721.0],"score":1.0,"text":"were used to give surfaces a natural-looking appearance by"},{"category_id":15,"poly":[864.0,1723.0,1566.0,1723.0,1566.0,1755.0,864.0,1755.0],"score":0.98,"text":"adjusting their specularity, reflectance, and diffusion. The"},{"category_id":15,"poly":[864.0,1788.0,1568.0,1788.0,1568.0,1820.0,864.0,1820.0],"score":1.0,"text":"frame rate of 30 frames per second, and a duration of 4"},{"category_id":15,"poly":[862.0,1817.0,1568.0,1820.0,1568.0,1859.0,861.0,1856.0],"score":0.98,"text":"seconds. In addition to performing photorealistic rendering."},{"category_id":15,"poly":[864.0,1856.0,1568.0,1856.0,1568.0,1888.0,864.0,1888.0],"score":0.99,"text":"depth renders of both video sequences were also generated and"},{"category_id":15,"poly":[864.0,1888.0,1566.0,1888.0,1566.0,1920.0,864.0,1920.0],"score":0.98,"text":"converted to ground truth disparity for the stereo video. The"},{"category_id":15,"poly":[862.0,1920.0,1564.0,1920.0,1564.0,1952.0,862.0,1952.0],"score":0.99,"text":"video sequences and ground truth data have been made avail-"},{"category_id":15,"poly":[862.0,1950.0,1566.0,1953.0,1566.0,1985.0,862.0,1982.0],"score":0.99,"text":"able at http://mc2.unl.edu/current-research"},{"category_id":15,"poly":[866.0,1989.0,1566.0,1989.0,1566.0,2019.0,866.0,2019.0],"score":0.98,"text":"/ image-processing/. Figure 2 shows two sample frames"},{"category_id":15,"poly":[862.0,1755.0,1312.0,1755.0,1312.0,1788.0,862.0,1788.0],"score":0.97,"text":"video sequence has a resolution of "},{"category_id":15,"poly":[1448.0,1755.0,1566.0,1755.0,1566.0,1788.0,1448.0,1788.0],"score":0.99,"text":"pixels,a"},{"category_id":15,"poly":[889.0,197.0,1566.0,199.0,1566.0,238.0,889.0,236.0],"score":1.0,"text":"Once the first iteration of stereo matching is complete,"},{"category_id":15,"poly":[864.0,268.0,1566.0,268.0,1566.0,300.0,864.0,300.0],"score":0.99,"text":"subsequent iterations. This is done by penalizing disparities"},{"category_id":15,"poly":[864.0,302.0,1568.0,302.0,1568.0,335.0,864.0,335.0],"score":1.0,"text":"that deviate from their expected values. The penalty function"},{"category_id":15,"poly":[862.0,337.0,996.0,337.0,996.0,369.0,862.0,369.0],"score":0.97,"text":"is given by"},{"category_id":15,"poly":[864.0,236.0,1094.0,236.0,1094.0,268.0,864.0,268.0],"score":0.96,"text":"disparityestimates"},{"category_id":15,"poly":[1135.0,236.0,1568.0,236.0,1568.0,268.0,1135.0,268.0],"score":0.97,"text":" can be used to guide matching in"},{"category_id":15,"poly":[157.0,1366.0,839.0,1368.0,838.0,1407.0,157.0,1405.0],"score":1.0,"text":"Having performed temporal cost aggregation, matches are"},{"category_id":15,"poly":[134.0,1405.0,834.0,1405.0,834.0,1437.0,134.0,1437.0],"score":0.99,"text":"determined using the Winner-Takes-All (WTA) match selec-"},{"category_id":15,"poly":[132.0,1506.0,374.0,1506.0,374.0,1538.0,132.0,1538.0],"score":1.0,"text":"cost, and is given by"},{"category_id":15,"poly":[692.0,1439.0,834.0,1439.0,834.0,1471.0,692.0,1471.0],"score":0.99,"text":", is the can-"},{"category_id":15,"poly":[134.0,1474.0,276.0,1474.0,276.0,1506.0,134.0,1506.0],"score":0.98,"text":"didate pixel"},{"category_id":15,"poly":[362.0,1474.0,836.0,1474.0,836.0,1506.0,362.0,1506.0],"score":0.99,"text":" characterized by the minimum matching"},{"category_id":15,"poly":[134.0,1439.0,461.0,1439.0,461.0,1471.0,134.0,1471.0],"score":1.0,"text":"tion criteria. The match for"},{"category_id":15,"poly":[481.0,1439.0,628.0,1439.0,628.0,1471.0,481.0,1471.0],"score":0.96,"text":", denoted as"},{"category_id":15,"poly":[134.0,548.0,838.0,545.0,838.0,577.0,134.0,580.0],"score":0.99,"text":"aggregation routine is exectuted. At each time instance, the"},{"category_id":15,"poly":[134.0,614.0,834.0,614.0,834.0,646.0,134.0,646.0],"score":1.0,"text":"weighted summation of costs obtained in the previous frames."},{"category_id":15,"poly":[132.0,646.0,838.0,644.0,838.0,676.0,132.0,678.0],"score":1.0,"text":"During temporal aggregation, the auxiliary cost is merged with"},{"category_id":15,"poly":[132.0,678.0,675.0,681.0,674.0,713.0,132.0,710.0],"score":0.99,"text":"the cost obtained from the current frame using"},{"category_id":15,"poly":[134.0,580.0,549.0,580.0,549.0,612.0,134.0,612.0],"score":1.0,"text":"algorithm stores an auxiliary cost"},{"category_id":15,"poly":[649.0,580.0,841.0,580.0,841.0,612.0,649.0,612.0],"score":0.96,"text":"which holds a"},{"category_id":15,"poly":[157.0,445.0,425.0,442.0,425.0,481.0,157.0,484.0],"score":0.98,"text":" Once aggregated costs"},{"category_id":15,"poly":[513.0,445.0,838.0,442.0,838.0,481.0,513.0,484.0],"score":0.96,"text":" have been computed for all"},{"category_id":15,"poly":[132.0,481.0,207.0,481.0,207.0,513.0,132.0,513.0],"score":1.0,"text":"pixels"},{"category_id":15,"poly":[228.0,481.0,838.0,481.0,838.0,513.0,228.0,513.0],"score":0.97,"text":" in the reference image and their respective matching"},{"category_id":15,"poly":[134.0,516.0,265.0,516.0,265.0,548.0,134.0,548.0],"score":1.0,"text":"candidates"},{"category_id":15,"poly":[289.0,516.0,838.0,516.0,838.0,548.0,289.0,548.0],"score":0.98,"text":" in the target image, a single-pass temporal"},{"category_id":15,"poly":[132.0,1116.0,841.0,1116.0,841.0,1155.0,132.0,1155.0],"score":0.99,"text":"in the temporal dimension. The temporal adaptive weight has "},{"category_id":15,"poly":[134.0,1153.0,838.0,1153.0,838.0,1185.0,134.0,1185.0],"score":0.99,"text":"the effect of preserving edges in the temporal domain, such"},{"category_id":15,"poly":[132.0,1182.0,836.0,1182.0,836.0,1215.0,132.0,1215.0],"score":0.98,"text":"that when a pixel coordinate transitions from one side of an"},{"category_id":15,"poly":[134.0,1219.0,838.0,1219.0,838.0,1251.0,134.0,1251.0],"score":0.98,"text":"edge to another in subsequent frames, the auxiliary cost is"},{"category_id":15,"poly":[134.0,1254.0,838.0,1254.0,838.0,1283.0,134.0,1283.0],"score":0.99,"text":"assigned a small weight and the majority of the cost is derived"},{"category_id":15,"poly":[130.0,1283.0,404.0,1286.0,404.0,1318.0,129.0,1315.0],"score":1.0,"text":"from the current frame."},{"category_id":15,"poly":[134.0,1086.0,207.0,1086.0,207.0,1118.0,134.0,1118.0],"score":0.99,"text":"where"},{"category_id":15,"poly":[237.0,1086.0,836.0,1086.0,836.0,1118.0,237.0,1118.0],"score":0.99,"text":"regulates the strength of grouping by color similarity"},{"category_id":15,"poly":[864.0,600.0,1568.0,600.0,1568.0,632.0,864.0,632.0],"score":1.0,"text":"and the matches are reselected using the WTA match selection"},{"category_id":15,"poly":[864.0,635.0,1568.0,635.0,1568.0,667.0,864.0,667.0],"score":0.99,"text":"criteria. The resulting disparity maps are then post-processed"},{"category_id":15,"poly":[864.0,669.0,1564.0,669.0,1564.0,699.0,864.0,699.0],"score":0.98,"text":"using a combination of median filtering and occlusion filling."},{"category_id":15,"poly":[864.0,701.0,1566.0,701.0,1566.0,731.0,864.0,731.0],"score":0.98,"text":"Finally, the current cost becomes the auxiliary cost for the next"},{"category_id":15,"poly":[862.0,731.0,1340.0,731.0,1340.0,770.0,862.0,770.0],"score":0.99,"text":"pair of frames in the video sequence, i.e.,"},{"category_id":15,"poly":[864.0,768.0,1017.0,768.0,1017.0,800.0,864.0,800.0],"score":1.0,"text":"for all pixels"},{"category_id":15,"poly":[1038.0,768.0,1465.0,768.0,1465.0,800.0,1038.0,800.0],"score":0.98,"text":" in the and their matching candidates"},{"category_id":15,"poly":[864.0,502.0,1427.0,502.0,1427.0,532.0,864.0,532.0],"score":1.0,"text":"values are incorporated into the matching cost as"},{"category_id":15,"poly":[864.0,468.0,1085.0,468.0,1085.0,500.0,864.0,500.0],"score":0.96,"text":"where the value of"},{"category_id":15,"poly":[1108.0,468.0,1564.0,468.0,1564.0,500.0,1108.0,500.0],"score":0.99,"text":"is chosen empirically. Next, the penalty"},{"category_id":15,"poly":[134.0,866.0,838.0,866.0,838.0,898.0,134.0,898.0],"score":0.99,"text":"temporal domain. The temporal adaptive weight computed"},{"category_id":15,"poly":[132.0,967.0,263.0,967.0,263.0,999.0,132.0,999.0],"score":0.93,"text":"is given by"},{"category_id":15,"poly":[134.0,834.0,320.0,834.0,320.0,866.0,134.0,866.0],"score":0.97,"text":"smoothing and"},{"category_id":15,"poly":[444.0,834.0,836.0,834.0,836.0,866.0,444.0,866.0],"score":0.92,"text":" enforces color similarity in the"},{"category_id":15,"poly":[178.0,930.0,838.0,928.0,839.0,967.0,178.0,969.0],"score":0.99,"text":", located at the same spatial coordinate in the prior frame,"},{"category_id":15,"poly":[132.0,795.0,490.0,800.0,490.0,832.0,132.0,827.0],"score":0.99,"text":"where the feedback coefficient"},{"category_id":15,"poly":[512.0,795.0,836.0,800.0,836.0,832.0,512.0,827.0],"score":0.97,"text":" controls the amount of cost"},{"category_id":15,"poly":[136.0,898.0,465.0,898.0,465.0,930.0,136.0,930.0],"score":0.99,"text":"between the pixel of interest"},{"category_id":15,"poly":[486.0,898.0,838.0,898.0,838.0,930.0,486.0,930.0],"score":1.0,"text":"in the current frame and pixel"},{"category_id":15,"poly":[159.0,1616.0,836.0,1616.0,836.0,1648.0,159.0,1648.0],"score":0.99,"text":"To asses the level of confidence associated with selecting"},{"category_id":15,"poly":[132.0,1648.0,836.0,1650.0,836.0,1682.0,132.0,1680.0],"score":1.0,"text":"minimum cost matches, the algorithm determines another set"},{"category_id":15,"poly":[134.0,1684.0,838.0,1684.0,838.0,1716.0,134.0,1716.0],"score":1.0,"text":"of matches, this time from the target to reference image, and"},{"category_id":15,"poly":[134.0,1783.0,182.0,1783.0,182.0,1815.0,134.0,1815.0],"score":1.0,"text":"and"},{"category_id":15,"poly":[136.0,1714.0,580.0,1714.0,580.0,1746.0,136.0,1746.0],"score":0.98,"text":"verifies if the results agree. Given that"},{"category_id":15,"poly":[305.0,1783.0,592.0,1783.0,592.0,1815.0,305.0,1815.0],"score":0.99,"text":", the confidence measure"},{"category_id":15,"poly":[628.0,1783.0,811.0,1783.0,811.0,1815.0,628.0,1815.0],"score":0.97,"text":"is computed as"},{"category_id":15,"poly":[132.0,1746.0,607.0,1751.0,607.0,1783.0,132.0,1778.0],"score":1.0,"text":"in the right image is the match for pixel"},{"category_id":15,"poly":[628.0,1746.0,836.0,1751.0,836.0,1783.0,628.0,1778.0],"score":0.98,"text":"in the left image,"},{"category_id":15,"poly":[695.0,1714.0,815.0,1714.0,815.0,1746.0,695.0,1746.0],"score":0.99,"text":", i.e. pixel"},{"category_id":15,"poly":[1132.0,814.0,1298.0,814.0,1298.0,852.0,1132.0,852.0],"score":1.0,"text":"IV. RESULTS"},{"category_id":15,"poly":[155.0,401.0,481.0,406.0,480.0,445.0,155.0,440.0],"score":0.99,"text":"Temporal cost aggregation"},{"category_id":15,"poly":[129.0,1325.0,718.0,1327.0,718.0,1366.0,129.0,1363.0],"score":0.99,"text":"C. Disparity Selection and Confidence Assessment"},{"category_id":15,"poly":[888.0,158.0,1252.0,158.0,1252.0,197.0,888.0,197.0],"score":0.97,"text":"Iterative Disparity Refinement"}],"page_info":{"page_no":2,"height":2200,"width":1700}},{"layout_dets":[{"category_id":1,"poly":[133.2669677734375,156.7020721435547,840.6729125976562,156.7020721435547,840.6729125976562,257.75836181640625,133.2669677734375,257.75836181640625],"score":0.9999951124191284},{"category_id":3,"poly":[866.177734375,171.2958526611328,1510.944580078125,171.2958526611328,1510.944580078125,848.8190307617188,866.177734375,848.8190307617188],"score":0.9999942779541016},{"category_id":1,"poly":[131.3756561279297,1520.5887451171875,838.545166015625,1520.5887451171875,838.545166015625,1885.353515625,131.3756561279297,1885.353515625],"score":0.9999925494194031},{"category_id":4,"poly":[131.56919860839844,1352.6187744140625,840.1758422851562,1352.6187744140625,840.1758422851562,1490.513671875,131.56919860839844,1490.513671875],"score":0.9999915361404419},{"category_id":1,"poly":[132.41786193847656,1886.0615234375,838.675537109375,1886.0615234375,838.675537109375,2019.347412109375,132.41786193847656,2019.347412109375],"score":0.9999526739120483},{"category_id":3,"poly":[136.71240234375,278.259765625,816.1984252929688,278.259765625,816.1984252929688,1348.5758056640625,136.71240234375,1348.5758056640625],"score":0.9999439120292664},{"category_id":1,"poly":[863.4852905273438,1917.056884765625,1569.6337890625,1917.056884765625,1569.6337890625,2020.57421875,863.4852905273438,2020.57421875],"score":0.9999344348907471},{"category_id":4,"poly":[861.7813720703125,1749.4459228515625,1567.659912109375,1749.4459228515625,1567.659912109375,1852.389892578125,861.7813720703125,1852.389892578125],"score":0.9986151456832886},{"category_id":3,"poly":[874.6467895507812,1536.7642822265625,1506.6514892578125,1536.7642822265625,1506.6514892578125,1734.9659423828125,874.6467895507812,1734.9659423828125],"score":0.9940656423568726},{"category_id":4,"poly":[859.3250122070312,861.2320556640625,1569.650634765625,861.2320556640625,1569.650634765625,1033.0804443359375,859.3250122070312,1033.0804443359375],"score":0.985899806022644},{"category_id":1,"poly":[861.6172485351562,1064.186279296875,1564.036865234375,1064.186279296875,1564.036865234375,1135.5125732421875,861.6172485351562,1135.5125732421875],"score":0.9128350019454956},{"category_id":3,"poly":[888.8074340820312,1163.7965087890625,1529.8028564453125,1163.7965087890625,1529.8028564453125,1510.91162109375,888.8074340820312,1510.91162109375],"score":0.7896175384521484},{"category_id":5,"poly":[900.75146484375,1161.0631103515625,1527.15673828125,1161.0631103515625,1527.15673828125,1490.2149658203125,900.75146484375,1490.2149658203125],"score":0.7772396802902222},{"category_id":0,"poly":[1178.85791015625,152.25347900390625,1284.6339111328125,152.25347900390625,1284.6339111328125,179.1011962890625,1178.85791015625,179.1011962890625],"score":0.5732811689376831},{"category_id":4,"poly":[1178.981689453125,152.21678161621094,1284.4158935546875,152.21678161621094,1284.4158935546875,179.05447387695312,1178.981689453125,179.05447387695312],"score":0.4503781795501709},{"category_id":13,"poly":[1295,896,1483,896,1483,931,1295,931],"score":0.93,"latex":"\\{\\pm0,\\pm20,\\pm40\\}"},{"category_id":13,"poly":[481,1919,534,1919,534,1949,481,1949],"score":0.87,"latex":"\\pm20"},{"category_id":13,"poly":[591,1919,644,1919,644,1949,591,1949],"score":0.87,"latex":"\\pm40"},{"category_id":13,"poly":[1227,1436,1253,1436,1253,1459,1227,1459],"score":0.86,"latex":"\\gamma_{c}"},{"category_id":13,"poly":[1295,1436,1323,1436,1323,1461,1295,1461],"score":0.85,"latex":"\\gamma_{g}"},{"category_id":13,"poly":[133,1588,186,1588,186,1618,133,1618],"score":0.85,"latex":"\\pm20"},{"category_id":13,"poly":[249,1587,302,1587,302,1618,249,1618],"score":0.84,"latex":"\\pm40"},{"category_id":13,"poly":[787,1555,828,1555,828,1585,787,1585],"score":0.82,"latex":"\\pm0"},{"category_id":13,"poly":[532,1421,572,1421,572,1452,532,1452],"score":0.81,"latex":"3^{\\mathrm{rd}}"},{"category_id":13,"poly":[230,1389,266,1389,266,1419,230,1419],"score":0.8,"latex":"1^{\\mathrm{st}}"},{"category_id":13,"poly":[655,1986,675,1986,675,2013,655,2013],"score":0.78,"latex":"\\lambda"},{"category_id":13,"poly":[200,1455,240,1455,240,1486,200,1486],"score":0.75,"latex":"4^{\\mathrm{th}}"},{"category_id":13,"poly":[954,1255,980,1255,980,1275,954,1275],"score":0.75,"latex":"\\gamma_{c}"},{"category_id":13,"poly":[954,1281,980,1281,980,1302,954,1302],"score":0.74,"latex":"\\gamma_{g}"},{"category_id":13,"poly":[959,1227,976,1227,976,1245,959,1245],"score":0.74,"latex":"\\tau"},{"category_id":13,"poly":[960,1352,976,1352,976,1372,960,1372],"score":0.72,"latex":"k"},{"category_id":13,"poly":[410,1986,430,1986,430,2013,410,2013],"score":0.7,"latex":"\\lambda"},{"category_id":13,"poly":[955,1331,979,1331,979,1351,955,1351],"score":0.7,"latex":"\\gamma_{t}"},{"category_id":13,"poly":[1489,1752,1510,1752,1510,1778,1489,1778],"score":0.69,"latex":"\\lambda"},{"category_id":13,"poly":[1176,965,1195,965,1195,992,1176,992],"score":0.69,"latex":"\\lambda"},{"category_id":13,"poly":[246,1421,289,1421,289,1452,246,1452],"score":0.69,"latex":"2^{\\mathrm{nd}}"},{"category_id":13,"poly":[958,1302,977,1302,977,1323,958,1323],"score":0.63,"latex":"\\lambda"},{"category_id":13,"poly":[959,1380,977,1380,977,1397,959,1397],"score":0.58,"latex":"\\alpha"},{"category_id":13,"poly":[436,1621,455,1621,455,1648,436,1648],"score":0.58,"latex":"\\lambda"},{"category_id":13,"poly":[959,1204,977,1204,977,1219,959,1219],"score":0.42,"latex":"\\omega"},{"category_id":13,"poly":[870,1592,890,1592,890,1617,870,1617],"score":0.31,"latex":"\\lambda"},{"category_id":15,"poly":[134.0,160.0,836.0,160.0,836.0,192.0,134.0,192.0],"score":0.99,"text":"of the synthetic stereo scene from a single camera perspective,"},{"category_id":15,"poly":[134.0,195.0,838.0,195.0,838.0,227.0,134.0,227.0],"score":0.99,"text":"along with the ground truth disparity, occlusion map, and"},{"category_id":15,"poly":[130.0,222.0,347.0,230.0,346.0,264.0,129.0,256.0],"score":0.99,"text":"discontinuity map."},{"category_id":15,"poly":[155.0,1517.0,841.0,1519.0,841.0,1558.0,155.0,1556.0],"score":0.99,"text":" The results of temporal stereo matching are given in Figure"},{"category_id":15,"poly":[132.0,1657.0,838.0,1657.0,838.0,1689.0,132.0,1689.0],"score":0.99,"text":"stereo matching methods, improvements are negligible when"},{"category_id":15,"poly":[132.0,1691.0,838.0,1691.0,838.0,1723.0,132.0,1723.0],"score":0.99,"text":"no noise is added to the images [10], [19]. This is largely due"},{"category_id":15,"poly":[132.0,1723.0,836.0,1723.0,836.0,1753.0,132.0,1753.0],"score":0.98,"text":"to the fact that the video used to evaluate these methods is"},{"category_id":15,"poly":[129.0,1753.0,838.0,1751.0,839.0,1790.0,129.0,1792.0],"score":0.99,"text":" computer generated with very little noise to start with, thus"},{"category_id":15,"poly":[134.0,1790.0,836.0,1790.0,836.0,1822.0,134.0,1822.0],"score":0.99,"text":"the noise suppression achieved with temporal stereo matching"},{"category_id":15,"poly":[132.0,1817.0,839.0,1822.0,838.0,1859.0,132.0,1854.0],"score":0.99,"text":"shows little to no improvement over methods that operate on"},{"category_id":15,"poly":[130.0,1856.0,319.0,1859.0,318.0,1891.0,129.0,1888.0],"score":0.99,"text":"pairs of images."},{"category_id":15,"poly":[187.0,1590.0,248.0,1590.0,248.0,1622.0,187.0,1622.0],"score":0.87,"text":",and"},{"category_id":15,"poly":[303.0,1590.0,838.0,1590.0,838.0,1622.0,303.0,1622.0],"score":0.98,"text":". Each performance plot is given as a function"},{"category_id":15,"poly":[127.0,1551.0,786.0,1554.0,786.0,1593.0,127.0,1590.0],"score":0.98,"text":" 3 for uniform additive noise confined to the ranges of"},{"category_id":15,"poly":[134.0,1622.0,435.0,1622.0,435.0,1655.0,134.0,1655.0],"score":0.99,"text":"of the feedback coefficient"},{"category_id":15,"poly":[456.0,1622.0,836.0,1622.0,836.0,1655.0,456.0,1655.0],"score":0.97,"text":". As with the majority of temporal"},{"category_id":15,"poly":[134.0,1359.0,834.0,1359.0,834.0,1391.0,134.0,1391.0],"score":0.99,"text":"Figure 2: Two sample frames from the synthetic video se-"},{"category_id":15,"poly":[573.0,1418.0,836.0,1421.0,836.0,1460.0,573.0,1457.0],"score":1.0,"text":"row), and discontinuity"},{"category_id":15,"poly":[134.0,1393.0,229.0,1393.0,229.0,1425.0,134.0,1425.0],"score":0.96,"text":"quence ("},{"category_id":15,"poly":[267.0,1393.0,836.0,1393.0,836.0,1425.0,267.0,1425.0],"score":0.98,"text":"row), along with their corresponding ground truth"},{"category_id":15,"poly":[127.0,1456.0,199.0,1450.0,199.0,1489.0,128.0,1495.0],"score":0.91,"text":"map ("},{"category_id":15,"poly":[241.0,1456.0,309.0,1450.0,310.0,1489.0,241.0,1495.0],"score":1.0,"text":"row)."},{"category_id":15,"poly":[129.0,1418.0,245.0,1421.0,245.0,1460.0,129.0,1457.0],"score":0.93,"text":" disparity "},{"category_id":15,"poly":[290.0,1418.0,531.0,1421.0,531.0,1460.0,290.0,1457.0],"score":1.0,"text":"row), occlusion map ("},{"category_id":15,"poly":[159.0,1888.0,836.0,1888.0,836.0,1920.0,159.0,1920.0],"score":0.99,"text":" Significant improvements in accuracy can be seen in Figure"},{"category_id":15,"poly":[132.0,1950.0,839.0,1955.0,838.0,1987.0,132.0,1982.0],"score":1.0,"text":"the effect of noise in the current frame is reduced by increasing"},{"category_id":15,"poly":[134.0,1920.0,480.0,1920.0,480.0,1952.0,134.0,1952.0],"score":0.99,"text":"3 when the noise has ranges of"},{"category_id":15,"poly":[535.0,1920.0,590.0,1920.0,590.0,1952.0,535.0,1952.0],"score":0.92,"text":" and"},{"category_id":15,"poly":[645.0,1920.0,836.0,1920.0,836.0,1952.0,645.0,1952.0],"score":0.96,"text":". In this scenario,"},{"category_id":15,"poly":[676.0,1989.0,838.0,1989.0,838.0,2019.0,676.0,2019.0],"score":0.98,"text":"has the effect"},{"category_id":15,"poly":[134.0,1989.0,409.0,1989.0,409.0,2019.0,134.0,2019.0],"score":1.0,"text":"the feedback coefficient"},{"category_id":15,"poly":[431.0,1989.0,654.0,1989.0,654.0,2019.0,431.0,2019.0],"score":0.97,"text":". This increasing of"},{"category_id":15,"poly":[864.0,1920.0,1566.0,1920.0,1566.0,1952.0,864.0,1952.0],"score":0.98,"text":"of averaging out noise in the per-pixel costs by selecting"},{"category_id":15,"poly":[861.0,1950.0,1566.0,1948.0,1566.0,1987.0,862.0,1989.0],"score":0.98,"text":"matches based more heavily upon the auxiliary cost, which"},{"category_id":15,"poly":[862.0,1989.0,1568.0,1989.0,1568.0,2021.0,862.0,2021.0],"score":0.99,"text":"is essentially a much more stable running average of the cost"},{"category_id":15,"poly":[864.0,1788.0,1564.0,1785.0,1564.0,1817.0,864.0,1820.0],"score":0.99,"text":"responding to the smallest mean squared error (MSE) of the"},{"category_id":15,"poly":[864.0,1822.0,1427.0,1822.0,1427.0,1854.0,864.0,1854.0],"score":0.99,"text":"disparity estimates for a range of noise strengths."},{"category_id":15,"poly":[862.0,1748.0,1488.0,1753.0,1488.0,1785.0,861.0,1781.0],"score":0.99,"text":"Figure 4: Optimal values of the feedback coefficient "},{"category_id":15,"poly":[1511.0,1748.0,1561.0,1753.0,1561.0,1785.0,1511.0,1781.0],"score":0.96,"text":"cor-"},{"category_id":15,"poly":[864.0,866.0,1566.0,866.0,1566.0,898.0,864.0,898.0],"score":0.99,"text":"Figure 3: Performance of temporal matching at different levels"},{"category_id":15,"poly":[864.0,935.0,1566.0,933.0,1566.0,965.0,864.0,967.0],"score":0.98,"text":"squared error (MSE) of disparities is plotted versus the values"},{"category_id":15,"poly":[864.0,1001.0,1492.0,1001.0,1492.0,1031.0,864.0,1031.0],"score":0.99,"text":"values of MSE obtained without temporal aggregation."},{"category_id":15,"poly":[864.0,901.0,1294.0,901.0,1294.0,933.0,864.0,933.0],"score":0.99,"text":"of uniformly distributed image noise"},{"category_id":15,"poly":[1484.0,901.0,1568.0,901.0,1568.0,933.0,1484.0,933.0],"score":0.99,"text":".Mean"},{"category_id":15,"poly":[864.0,967.0,1175.0,967.0,1175.0,999.0,864.0,999.0],"score":0.99,"text":"of the feedback coefficient"},{"category_id":15,"poly":[1196.0,967.0,1568.0,967.0,1568.0,999.0,1196.0,999.0],"score":0.99,"text":". Dashed lines correspond to the"},{"category_id":15,"poly":[857.0,1061.0,1566.0,1068.0,1566.0,1107.0,857.0,1100.0],"score":0.99,"text":" Table I: Parameters used in the evaluation of real-time tempo-"},{"category_id":15,"poly":[859.0,1102.0,1093.0,1105.0,1092.0,1137.0,859.0,1134.0],"score":1.0,"text":"ral stereo matching."},{"category_id":15,"poly":[1178.0,151.0,1282.0,151.0,1282.0,186.0,1178.0,186.0],"score":1.0,"text":"Noise: \u00b10"},{"category_id":15,"poly":[1178.0,151.0,1282.0,151.0,1282.0,186.0,1178.0,186.0],"score":1.0,"text":"Noise: \u00b10"}],"page_info":{"page_no":3,"height":2200,"width":1700}},{"layout_dets":[{"category_id":5,"poly":[880.81298828125,613.750244140625,1552.5638427734375,613.750244140625,1552.5638427734375,855.9174194335938,880.81298828125,855.9174194335938],"score":0.9999957084655762},{"category_id":1,"poly":[862.7925415039062,158.05548095703125,1569.6671142578125,158.05548095703125,1569.6671142578125,456.6153869628906,862.7925415039062,456.6153869628906],"score":0.9999922513961792},{"category_id":1,"poly":[864.6585083007812,1061.7374267578125,1570.4825439453125,1061.7374267578125,1570.4825439453125,1459.7132568359375,864.6585083007812,1459.7132568359375],"score":0.9999921321868896},{"category_id":1,"poly":[130.64285278320312,1519.7022705078125,836.2221069335938,1519.7022705078125,836.2221069335938,1882.68359375,130.64285278320312,1882.68359375],"score":0.9999898672103882},{"category_id":1,"poly":[133.1135711669922,158.4307861328125,837.9683837890625,158.4307861328125,837.9683837890625,323.343017578125,133.1135711669922,323.343017578125],"score":0.9999892115592957},{"category_id":4,"poly":[132.3511199951172,1347.8763427734375,839.7514038085938,1347.8763427734375,839.7514038085938,1476.9757080078125,132.3511199951172,1476.9757080078125],"score":0.9999880790710449},{"category_id":7,"poly":[887.6280517578125,860.9362182617188,1551.5972900390625,860.9362182617188,1551.5972900390625,964.0142211914062,887.6280517578125,964.0142211914062],"score":0.9999836683273315},{"category_id":1,"poly":[869.9986572265625,1514.7762451171875,1571.624755859375,1514.7762451171875,1571.624755859375,2022.618896484375,869.9986572265625,2022.618896484375],"score":0.9999811053276062},{"category_id":3,"poly":[164.82151794433594,352.74810791015625,805.8219604492188,352.74810791015625,805.8219604492188,1320.43310546875,164.82151794433594,1320.43310546875],"score":0.9999799728393555},{"category_id":0,"poly":[1137.668701171875,1477.0120849609375,1293.498046875,1477.0120849609375,1293.498046875,1502.5439453125,1137.668701171875,1502.5439453125],"score":0.9999679327011108},{"category_id":1,"poly":[133.0285186767578,1886.7501220703125,837.0147705078125,1886.7501220703125,837.0147705078125,2018.0294189453125,133.0285186767578,2018.0294189453125],"score":0.9999630451202393},{"category_id":0,"poly":[1114.8399658203125,1022.4933471679688,1317.0313720703125,1022.4933471679688,1317.0313720703125,1052.679931640625,1114.8399658203125,1052.679931640625],"score":0.9999338984489441},{"category_id":1,"poly":[862.0576171875,480.8196105957031,1565.8367919921875,480.8196105957031,1565.8367919921875,577.5508422851562,862.0576171875,577.5508422851562],"score":0.8958550691604614},{"category_id":6,"poly":[862.0606079101562,480.7809753417969,1565.667724609375,480.7809753417969,1565.667724609375,577.4689331054688,862.0606079101562,577.4689331054688],"score":0.4145430028438568},{"category_id":13,"poly":[736,1445,827,1445,827,1475,736,1475],"score":0.9,"latex":"\\lambda=0.8"},{"category_id":13,"poly":[1003,887,1105,887,1105,911,1003,911],"score":0.89,"latex":"320\\times240"},{"category_id":13,"poly":[338,1446,391,1446,391,1475,338,1475],"score":0.87,"latex":"\\pm30"},{"category_id":13,"poly":[166,1619,219,1619,219,1649,166,1649],"score":0.85,"latex":"\\pm40"},{"category_id":13,"poly":[301,196,329,196,329,224,301,224],"score":0.84,"latex":"\\gamma_{t}"},{"category_id":13,"poly":[795,1586,836,1586,836,1616,795,1616],"score":0.84,"latex":"\\pm0"},{"category_id":13,"poly":[1037,939,1059,939,1059,960,1037,960],"score":0.83,"latex":"\\%"},{"category_id":13,"poly":[462,1586,482,1586,482,1613,462,1613],"score":0.78,"latex":"\\lambda"},{"category_id":15,"poly":[862.0,160.0,1571.0,160.0,1571.0,192.0,862.0,192.0],"score":0.98,"text":"the proposed implementation achieves the highest speed of"},{"category_id":15,"poly":[864.0,195.0,1566.0,195.0,1566.0,227.0,864.0,227.0],"score":0.99,"text":"operation measured by the number of disparity hypotheses"},{"category_id":15,"poly":[864.0,227.0,1568.0,227.0,1568.0,259.0,864.0,259.0],"score":0.99,"text":"evaluated per second, as shown in Table I1. It is also the second"},{"category_id":15,"poly":[862.0,261.0,1568.0,261.0,1568.0,293.0,862.0,293.0],"score":0.99,"text":"most accurate real-time method in terms of error rate, as"},{"category_id":15,"poly":[864.0,296.0,1564.0,296.0,1564.0,325.0,864.0,325.0],"score":1.0,"text":"measured using the Middlebury stereo evaluation benchmark."},{"category_id":15,"poly":[859.0,323.0,1568.0,325.0,1568.0,358.0,859.0,355.0],"score":0.98,"text":" It should be noted that it is difficult to establish an unbiased"},{"category_id":15,"poly":[862.0,358.0,1566.0,358.0,1566.0,390.0,862.0,390.0],"score":1.0,"text":"metric for speed comparisons, as the architecture, number of"},{"category_id":15,"poly":[866.0,394.0,1568.0,394.0,1568.0,426.0,866.0,426.0],"score":0.98,"text":"cores, and clock speed of graphics hardware used are not"},{"category_id":15,"poly":[862.0,424.0,1259.0,429.0,1259.0,461.0,861.0,456.0],"score":0.99,"text":"consistent across implementations."},{"category_id":15,"poly":[889.0,1061.0,1571.0,1061.0,1571.0,1100.0,889.0,1100.0],"score":1.0,"text":"While the majority of stereo matching algorithms focus"},{"category_id":15,"poly":[859.0,1093.0,1571.0,1095.0,1571.0,1134.0,859.0,1132.0],"score":0.99,"text":" on achieving high accuracy on still images, the volume of"},{"category_id":15,"poly":[862.0,1130.0,1564.0,1130.0,1564.0,1162.0,862.0,1162.0],"score":0.99,"text":"research aimed at recovery of temporally consistent disparity"},{"category_id":15,"poly":[862.0,1162.0,1568.0,1162.0,1568.0,1201.0,862.0,1201.0],"score":0.99,"text":"maps remains disproportionally small. This paper introduces"},{"category_id":15,"poly":[862.0,1196.0,1568.0,1196.0,1568.0,1235.0,862.0,1235.0],"score":0.98,"text":"an efficient temporal cost aggregation scheme that can easily"},{"category_id":15,"poly":[859.0,1226.0,1571.0,1228.0,1571.0,1267.0,859.0,1265.0],"score":0.99,"text":"be combined with conventional spatial cost aggregation to"},{"category_id":15,"poly":[864.0,1265.0,1568.0,1265.0,1568.0,1297.0,864.0,1297.0],"score":1.0,"text":"improve the accuracy of stereo matching when operating on"},{"category_id":15,"poly":[864.0,1297.0,1568.0,1297.0,1568.0,1329.0,864.0,1329.0],"score":0.99,"text":"video sequences. A synthetic video sequence, along with"},{"category_id":15,"poly":[864.0,1331.0,1568.0,1331.0,1568.0,1364.0,864.0,1364.0],"score":0.99,"text":"ground truth disparity data, was generated to evaluate the"},{"category_id":15,"poly":[862.0,1361.0,1571.0,1361.0,1571.0,1400.0,862.0,1400.0],"score":0.98,"text":"performance of the proposed method. It was shown that"},{"category_id":15,"poly":[864.0,1398.0,1571.0,1398.0,1571.0,1430.0,864.0,1430.0],"score":0.98,"text":"temporal aggregation is significantly more robust to noise than"},{"category_id":15,"poly":[862.0,1430.0,1497.0,1430.0,1497.0,1462.0,862.0,1462.0],"score":0.99,"text":"a method that only considers the current stereo frames."},{"category_id":15,"poly":[157.0,1517.0,838.0,1517.0,838.0,1556.0,157.0,1556.0],"score":0.99,"text":"The optimal value of the feedback coefficient is largely"},{"category_id":15,"poly":[134.0,1554.0,836.0,1554.0,836.0,1584.0,134.0,1584.0],"score":0.97,"text":"dependent on the noise being added to the image. Figure 4"},{"category_id":15,"poly":[132.0,1655.0,838.0,1655.0,838.0,1684.0,132.0,1684.0],"score":0.99,"text":"rely on the auxiliary cost when noise is high and it is more"},{"category_id":15,"poly":[132.0,1684.0,839.0,1689.0,838.0,1721.0,132.0,1716.0],"score":0.98,"text":"beneficial to rely on the current cost when noise is low. Figure"},{"category_id":15,"poly":[132.0,1719.0,839.0,1723.0,838.0,1755.0,132.0,1751.0],"score":1.0,"text":"5 illustrates the improvements that are achieved when applying"},{"category_id":15,"poly":[134.0,1755.0,836.0,1755.0,836.0,1785.0,134.0,1785.0],"score":0.98,"text":"temporal stereo matching to a particular pair of frames in the"},{"category_id":15,"poly":[134.0,1788.0,834.0,1788.0,834.0,1820.0,134.0,1820.0],"score":1.0,"text":"synthetic video sequence. Clearly, the noise in the disparity"},{"category_id":15,"poly":[134.0,1822.0,836.0,1822.0,836.0,1854.0,134.0,1854.0],"score":0.99,"text":"map is drastically reduced when temporal stereo matching is"},{"category_id":15,"poly":[132.0,1856.0,196.0,1856.0,196.0,1886.0,132.0,1886.0],"score":1.0,"text":"used."},{"category_id":15,"poly":[132.0,1620.0,165.0,1620.0,165.0,1652.0,132.0,1652.0],"score":0.99,"text":"to"},{"category_id":15,"poly":[220.0,1620.0,838.0,1620.0,838.0,1652.0,220.0,1652.0],"score":0.98,"text":". As intuition would suggest, it is more beneficial to"},{"category_id":15,"poly":[127.0,1584.0,461.0,1581.0,461.0,1620.0,127.0,1623.0],"score":0.96,"text":" shows the optimal values of"},{"category_id":15,"poly":[483.0,1584.0,794.0,1581.0,794.0,1620.0,483.0,1623.0],"score":0.99,"text":"for noise ranging between"},{"category_id":15,"poly":[134.0,160.0,836.0,160.0,836.0,192.0,134.0,192.0],"score":0.99,"text":"over the most recent frames. By maintaining a reasonably"},{"category_id":15,"poly":[134.0,229.0,836.0,229.0,836.0,261.0,134.0,261.0],"score":0.98,"text":"edges, essentially reducing over-smoothing of a pixel's dis-"},{"category_id":15,"poly":[132.0,261.0,838.0,261.0,838.0,293.0,132.0,293.0],"score":0.99,"text":"parity when a pixel transitions from one depth to another in"},{"category_id":15,"poly":[130.0,293.0,354.0,296.0,353.0,328.0,129.0,325.0],"score":1.0,"text":"subsequent frames."},{"category_id":15,"poly":[134.0,192.0,300.0,192.0,300.0,225.0,134.0,225.0],"score":0.93,"text":"high value of"},{"category_id":15,"poly":[330.0,192.0,836.0,192.0,836.0,225.0,330.0,225.0],"score":0.99,"text":", the auxiliary cost also preserves temporal"},{"category_id":15,"poly":[132.0,1345.0,836.0,1348.0,836.0,1382.0,132.0,1380.0],"score":1.0,"text":"Figure 5: A comparison of stereo matching without temporal"},{"category_id":15,"poly":[132.0,1382.0,834.0,1382.0,834.0,1414.0,132.0,1414.0],"score":0.98,"text":"cost aggregation (top\uff09 and with temporal cost aggregation"},{"category_id":15,"poly":[134.0,1416.0,836.0,1416.0,836.0,1446.0,134.0,1446.0],"score":0.98,"text":"(bottom) for a single frame in the synthetic video sequence"},{"category_id":15,"poly":[134.0,1448.0,337.0,1446.0,337.0,1478.0,134.0,1480.0],"score":0.98,"text":"where the noise is"},{"category_id":15,"poly":[392.0,1448.0,735.0,1446.0,735.0,1478.0,392.0,1480.0],"score":0.99,"text":"and the feedback coefficient is"},{"category_id":15,"poly":[896.0,855.0,1324.0,857.0,1323.0,896.0,896.0,894.0],"score":0.95,"text":"1I Millions of Disparity Estimates per Second."},{"category_id":15,"poly":[903.0,912.0,1550.0,912.0,1550.0,944.0,903.0,944.0],"score":0.99,"text":"3 As measured by the Middlebury stereo performance benchmark using"},{"category_id":15,"poly":[901.0,887.0,1002.0,887.0,1002.0,919.0,901.0,919.0],"score":0.99,"text":"2Assumes"},{"category_id":15,"poly":[1106.0,887.0,1404.0,887.0,1404.0,919.0,1106.0,919.0],"score":0.98,"text":"images with 32 disparity levels."},{"category_id":15,"poly":[915.0,937.0,1036.0,937.0,1036.0,969.0,915.0,969.0],"score":0.96,"text":"the avgerage"},{"category_id":15,"poly":[1060.0,937.0,1192.0,937.0,1192.0,969.0,1060.0,969.0],"score":0.96,"text":"of bad pixels."},{"category_id":15,"poly":[873.0,1515.0,1571.0,1515.0,1571.0,1545.0,873.0,1545.0],"score":0.97,"text":"[1] D. Scharstein and R. Szeliski, \u201cA taxonomy and evaluation of dense "},{"category_id":15,"poly":[915.0,1542.0,1573.0,1542.0,1573.0,1572.0,915.0,1572.0],"score":0.98,"text":"two-frame stereo correspondence algorithms\u201d\u2019 International Journal of"},{"category_id":15,"poly":[915.0,1565.0,1409.0,1565.0,1409.0,1597.0,915.0,1597.0],"score":0.98,"text":"Computer Vision, vol. 47, pp. 7-42, April-June 2002."},{"category_id":15,"poly":[871.0,1588.0,1568.0,1590.0,1568.0,1623.0,871.0,1620.0],"score":0.98,"text":"[2] D. Scharstein and R. Szeliski, \u201cHigh-accuracy stereo depth maps using"},{"category_id":15,"poly":[915.0,1616.0,1568.0,1616.0,1568.0,1648.0,915.0,1648.0],"score":0.97,"text":"structured light,\u201d in In IEEE Computer Society Conference on Computer"},{"category_id":15,"poly":[915.0,1641.0,1508.0,1641.0,1508.0,1673.0,915.0,1673.0],"score":0.98,"text":"Vision and Pattern Recognition, vol. 1, pp. 195-202, June 2003."},{"category_id":15,"poly":[873.0,1666.0,1568.0,1666.0,1568.0,1696.0,873.0,1696.0],"score":0.99,"text":"[3] J. Kowalczuk, E. Psota, and L. Perez, \u201cReal-time stereo matching on"},{"category_id":15,"poly":[912.0,1689.0,1571.0,1689.0,1571.0,1721.0,912.0,1721.0],"score":0.98,"text":" CUDA using an iterative refinement method for adaptive support-weight"},{"category_id":15,"poly":[915.0,1714.0,1571.0,1714.0,1571.0,1746.0,915.0,1746.0],"score":0.99,"text":"correspondences,\u201d Circuits and Systems for Video Technology, IEEE"},{"category_id":15,"poly":[908.0,1737.0,1374.0,1735.0,1374.0,1774.0,908.0,1776.0],"score":0.96,"text":"Transactions on, vol. 23, Ppp. 94 -104, Jan. 2013."},{"category_id":15,"poly":[873.0,1765.0,1568.0,1765.0,1568.0,1797.0,873.0,1797.0],"score":0.99,"text":"[4] K.-J. Yoon and I.-S. Kweon, Locally adaptive support-weight approach"},{"category_id":15,"poly":[912.0,1790.0,1571.0,1790.0,1571.0,1822.0,912.0,1822.0],"score":0.97,"text":"for visual correspondence search,' in CVPR'05: Proceedings of the 2005"},{"category_id":15,"poly":[915.0,1815.0,1571.0,1815.0,1571.0,1847.0,915.0,1847.0],"score":0.96,"text":"IEEE Computer Society Conference on ComputerVision andPattern"},{"category_id":15,"poly":[915.0,1840.0,1568.0,1840.0,1568.0,1872.0,915.0,1872.0],"score":0.97,"text":"Recognition (CVPR'05) - Volume 2, (Washington, DC, USA), Pp. 924-"},{"category_id":15,"poly":[912.0,1863.0,1247.0,1863.0,1247.0,1895.0,912.0,1895.0],"score":0.98,"text":"931, IEEE Computer Society, 2005."},{"category_id":15,"poly":[873.0,1891.0,1568.0,1891.0,1568.0,1923.0,873.0,1923.0],"score":0.97,"text":"[5] L. Wang, M. Liao, M. Gong, R. Yang, and D. Nister, \u201cHigh-quality real-"},{"category_id":15,"poly":[912.0,1916.0,1566.0,1916.0,1566.0,1946.0,912.0,1946.0],"score":0.99,"text":"time stereo using adaptive cost aggregation and dynamic programming,\""},{"category_id":15,"poly":[910.0,1936.0,1568.0,1939.0,1568.0,1971.0,910.0,1969.0],"score":0.94,"text":"in 3DPVT'06:Proceedings of the Third International Symposium"},{"category_id":15,"poly":[915.0,1964.0,1568.0,1964.0,1568.0,1996.0,915.0,1996.0],"score":0.98,"text":"on 3D Data Processing, Visualization, and Transmission (3DPVT'06),"},{"category_id":15,"poly":[915.0,1989.0,1564.0,1989.0,1564.0,2021.0,915.0,2021.0],"score":1.0,"text":"(Washington, DC, USA), Pp. 798-805, IEEE Computer Society, 2006."},{"category_id":15,"poly":[1134.0,1471.0,1296.0,1471.0,1296.0,1510.0,1134.0,1510.0],"score":1.0,"text":"REFERENCES"},{"category_id":15,"poly":[159.0,1888.0,836.0,1888.0,836.0,1920.0,159.0,1920.0],"score":0.99,"text":"The algorithm was implement using NVIDIA's Compute"},{"category_id":15,"poly":[134.0,1920.0,834.0,1920.0,834.0,1950.0,134.0,1950.0],"score":0.98,"text":"Unified Device Architecture (CUDA). The details of the im-"},{"category_id":15,"poly":[129.0,1948.0,841.0,1950.0,841.0,1989.0,129.0,1987.0],"score":0.98,"text":" plementation are similar to those given in [3]. When compared "},{"category_id":15,"poly":[132.0,1989.0,836.0,1989.0,836.0,2021.0,132.0,2021.0],"score":0.99,"text":"to other existing real-time stereo matching implementations,"},{"category_id":15,"poly":[1111.0,1022.0,1317.0,1022.0,1317.0,1061.0,1111.0,1061.0],"score":1.0,"text":"V. CONCLUSION"},{"category_id":15,"poly":[864.0,484.0,1564.0,484.0,1564.0,516.0,864.0,516.0],"score":0.99,"text":"Table II: A comparison of speed and accuracy for the imple-"},{"category_id":15,"poly":[864.0,518.0,1564.0,518.0,1564.0,550.0,864.0,550.0],"score":0.99,"text":"mentations of many leading real-time stereo matching meth-"},{"category_id":15,"poly":[862.0,550.0,917.0,550.0,917.0,584.0,862.0,584.0],"score":0.96,"text":"ods."},{"category_id":15,"poly":[864.0,484.0,1564.0,484.0,1564.0,516.0,864.0,516.0],"score":0.99,"text":"Table II: A comparison of speed and accuracy for the imple-"},{"category_id":15,"poly":[864.0,518.0,1564.0,518.0,1564.0,550.0,864.0,550.0],"score":0.99,"text":"mentations of many leading real-time stereo matching meth-"},{"category_id":15,"poly":[862.0,550.0,917.0,550.0,917.0,584.0,862.0,584.0],"score":0.96,"text":"ods."}],"page_info":{"page_no":4,"height":2200,"width":1700}},{"layout_dets":[{"category_id":1,"poly":[134.58497619628906,157.681884765625,841.3460693359375,157.681884765625,841.3460693359375,1666.27001953125,134.58497619628906,1666.27001953125],"score":0.9999936819076538},{"category_id":15,"poly":[143.0,163.0,838.0,163.0,838.0,192.0,143.0,192.0],"score":0.97,"text":"[6] W. Yu, T. Chen, F. Franchetti, and J. C. Hoe, \u201cHigh performance stereo"},{"category_id":15,"poly":[182.0,188.0,838.0,188.0,838.0,218.0,182.0,218.0],"score":0.98,"text":"vision designed for massively data parallel platforms,\u2019 Circuits and"},{"category_id":15,"poly":[182.0,213.0,841.0,213.0,841.0,245.0,182.0,245.0],"score":0.98,"text":"Systems for Video Technology, IEEE Transactions on, vol. 20, pp. 1509"},{"category_id":15,"poly":[182.0,238.0,411.0,238.0,411.0,268.0,182.0,268.0],"score":0.98,"text":"-1519, November 2010."},{"category_id":15,"poly":[143.0,264.0,838.0,264.0,838.0,293.0,143.0,293.0],"score":0.99,"text":"[7] S. Mattoccia, M. Viti, and F. Ries, \u201cNear real-time fast bilateral stereo"},{"category_id":15,"poly":[182.0,289.0,838.0,289.0,838.0,319.0,182.0,319.0],"score":0.96,"text":"on the GPU in Computer Vision and Pattern Recognition Workshops"},{"category_id":15,"poly":[178.0,307.0,841.0,309.0,841.0,348.0,178.0,346.0],"score":0.95,"text":"(CVPRW), 2011 IEEE Computer Society Conference on,Ppp. 136 -143,"},{"category_id":15,"poly":[185.0,339.0,289.0,339.0,289.0,364.0,185.0,364.0],"score":0.98,"text":"June 2011."},{"category_id":15,"poly":[141.0,362.0,838.0,362.0,838.0,392.0,141.0,392.0],"score":0.98,"text":"[8] K. Zhang, J. Lu, Q. Yang, G. Lafruit, R. Lauwereins, and L. Van Gool,"},{"category_id":15,"poly":[182.0,387.0,838.0,387.0,838.0,419.0,182.0,419.0],"score":0.98,"text":"\"Real-time and accurate stereo: A scalable approach with bitwise fast"},{"category_id":15,"poly":[185.0,412.0,838.0,412.0,838.0,445.0,185.0,445.0],"score":0.97,"text":"voting on CUDA,\u201d Circuits and Systems for Video Technology, IEEE"},{"category_id":15,"poly":[182.0,438.0,656.0,438.0,656.0,468.0,182.0,468.0],"score":0.99,"text":"Transactions on, vol. 21, pp. 867 -878, July 2011."},{"category_id":15,"poly":[141.0,463.0,838.0,463.0,838.0,493.0,141.0,493.0],"score":0.96,"text":"[9] C. Rhemann, A. Hosni, M. Bleyer, C. Rother, and M. Gelautz, \u201cFast cost-"},{"category_id":15,"poly":[182.0,488.0,838.0,488.0,838.0,518.0,182.0,518.0],"score":0.98,"text":"volume filtering for visual correspondence and beyond,\" in Computer"},{"category_id":15,"poly":[180.0,509.0,841.0,511.0,841.0,543.0,180.0,541.0],"score":0.95,"text":"Vision and Pattern Recognition (CVPR), 20ll IEEE Conference on,"},{"category_id":15,"poly":[180.0,536.0,448.0,534.0,448.0,566.0,180.0,568.0],"score":0.99,"text":"Pp. 3017 -3024, June 2011."},{"category_id":15,"poly":[134.0,561.0,838.0,561.0,838.0,591.0,134.0,591.0],"score":0.99,"text":"[10] A. Hosni, C. Rhemann, M. Bleyer, and M. Gelautz, \u201cTemporally con-"},{"category_id":15,"poly":[180.0,587.0,836.0,587.0,836.0,616.0,180.0,616.0],"score":0.99,"text":" sistent disparity and optical flow via efficient spatio-temporal filtering,\""},{"category_id":15,"poly":[182.0,612.0,838.0,612.0,838.0,642.0,182.0,642.0],"score":0.97,"text":"in Advances in Image and Video Technology (Y.-S. Ho, ed.), vol. 7087"},{"category_id":15,"poly":[180.0,632.0,845.0,632.0,845.0,671.0,180.0,671.0],"score":0.88,"text":"of Lectureotes inComputer Science,pp.16517,Springererlin /"},{"category_id":15,"poly":[182.0,660.0,353.0,660.0,353.0,692.0,182.0,692.0],"score":1.0,"text":"Heidelberg, 2012."},{"category_id":15,"poly":[134.0,685.0,838.0,685.0,838.0,717.0,134.0,717.0],"score":0.98,"text":"[11] C. Tomasi and R. Manduchi, \u201cBilateral filtering for gray and color"},{"category_id":15,"poly":[182.0,710.0,838.0,710.0,838.0,742.0,182.0,742.0],"score":0.98,"text":"images,\u201d in Computer Vision, 1998. Sixth International Conference on,"},{"category_id":15,"poly":[180.0,736.0,411.0,731.0,411.0,763.0,181.0,768.0],"score":0.93,"text":"pPp. 839 -846, jan 1998."},{"category_id":15,"poly":[132.0,761.0,838.0,761.0,838.0,791.0,132.0,791.0],"score":0.97,"text":"[12] K. He, J. Sun, and X. Tang, \u201cGuided image filtering,\u201d\u2019 in Computer"},{"category_id":15,"poly":[180.0,784.0,838.0,786.0,838.0,818.0,180.0,816.0],"score":0.98,"text":"Vision - ECCV 2010, vol. 6311 of Lecture Notes in Computer Science,"},{"category_id":15,"poly":[180.0,811.0,607.0,807.0,608.0,839.0,180.0,843.0],"score":0.98,"text":"pp. 1-14, Springer Berlin / Heidelberg, 2010."},{"category_id":15,"poly":[129.0,832.0,839.0,837.0,838.0,869.0,129.0,864.0],"score":0.98,"text":"[13] L. Zhang, B. Curless, and S. M. Seitz, \u201cSpacetime stereo: Shape"},{"category_id":15,"poly":[182.0,862.0,836.0,862.0,836.0,891.0,182.0,891.0],"score":0.98,"text":"recovery for dynamic scenes,\u201d in IEEE Computer Society Conference"},{"category_id":15,"poly":[182.0,885.0,834.0,885.0,834.0,917.0,182.0,917.0],"score":0.97,"text":"on Computer Vision and Pattern Recognition, pp. 367-374, June 2003."},{"category_id":15,"poly":[132.0,910.0,838.0,910.0,838.0,940.0,132.0,940.0],"score":0.98,"text":"[14] J. Davis, D. Nehab, R. Ramamoorthi, and S. Rusinkiewicz, \u201cSpacetime"},{"category_id":15,"poly":[182.0,935.0,838.0,935.0,838.0,965.0,182.0,965.0],"score":0.97,"text":"stereo: a unifying framework for depth from triangulation,\u201d\u2019 Pattern"},{"category_id":15,"poly":[182.0,960.0,838.0,960.0,838.0,990.0,182.0,990.0],"score":0.98,"text":"Analysis and Machine Intelligence, IEEE Transactions on,vol. 27,"},{"category_id":15,"poly":[180.0,983.0,462.0,983.0,462.0,1015.0,180.0,1015.0],"score":0.97,"text":"Pp. 296 -302, February 2005."},{"category_id":15,"poly":[132.0,1011.0,838.0,1011.0,838.0,1040.0,132.0,1040.0],"score":0.99,"text":"[15] E. Larsen, P. Mordohai, M. Pollefeys, and H. Fuchs, \u201cTemporally"},{"category_id":15,"poly":[182.0,1036.0,836.0,1036.0,836.0,1066.0,182.0,1066.0],"score":0.99,"text":"consistent reconstruction from multiple video streams using enhanced"},{"category_id":15,"poly":[178.0,1054.0,843.0,1056.0,843.0,1095.0,178.0,1093.0],"score":0.95,"text":"belief propagation in Computer Vision, 2007.ICCV 2007. IEEE1lth"},{"category_id":15,"poly":[180.0,1082.0,644.0,1082.0,644.0,1121.0,180.0,1121.0],"score":0.97,"text":"International Conference on, pp. 1 -8, oct. 2007."},{"category_id":15,"poly":[134.0,1109.0,838.0,1109.0,838.0,1141.0,134.0,1141.0],"score":0.97,"text":"[16] M. Bleyer, M. Gelautz, C. Rother, and C. Rhemann, \u201c\"A stereo approach"},{"category_id":15,"poly":[180.0,1134.0,838.0,1134.0,838.0,1166.0,180.0,1166.0],"score":0.99,"text":"that handles the mating problem via image warping\" in Computer"},{"category_id":15,"poly":[182.0,1157.0,838.0,1157.0,838.0,1189.0,182.0,1189.0],"score":0.98,"text":"Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference"},{"category_id":15,"poly":[180.0,1183.0,459.0,1175.0,460.0,1212.0,181.0,1219.0],"score":0.98,"text":"on, pp. 501 -508, June 2009."},{"category_id":15,"poly":[129.0,1205.0,838.0,1208.0,838.0,1240.0,129.0,1237.0],"score":0.98,"text":" [17] M. Sizintsev and R. Wildes, \u201cSpatiotemporal stereo via spatiotemporal"},{"category_id":15,"poly":[182.0,1235.0,838.0,1235.0,838.0,1265.0,182.0,1265.0],"score":0.97,"text":"quadric element (stequel) matching,\u201d in Computer Vision and Pattern"},{"category_id":15,"poly":[185.0,1258.0,841.0,1258.0,841.0,1290.0,185.0,1290.0],"score":0.98,"text":"Recognition, 2009. CVPR 2009. IEEE Conference on, Pp. 493 -500,"},{"category_id":15,"poly":[185.0,1286.0,286.0,1286.0,286.0,1311.0,185.0,1311.0],"score":0.99,"text":"june 2009."},{"category_id":15,"poly":[132.0,1309.0,838.0,1309.0,838.0,1338.0,132.0,1338.0],"score":0.97,"text":"[18] M. Sizintsev and R. Wildes, \u201cSpatiotemporal stereo and scene flow via"},{"category_id":15,"poly":[182.0,1334.0,841.0,1334.0,841.0,1364.0,182.0,1364.0],"score":0.97,"text":"stequel matching,\u201d\u2019Pattern Analysis and Machine Intelligence, IEEE"},{"category_id":15,"poly":[182.0,1359.0,684.0,1359.0,684.0,1391.0,182.0,1391.0],"score":1.0,"text":"Transactions on, vol. 34, pp. 1206 -1219, june 2012."},{"category_id":15,"poly":[132.0,1382.0,834.0,1382.0,834.0,1412.0,132.0,1412.0],"score":0.98,"text":"[19] C. Richardt, D. Orr, I. Davies, A. Criminisi, and N. A. Dodgson,"},{"category_id":15,"poly":[185.0,1409.0,838.0,1409.0,838.0,1441.0,185.0,1441.0],"score":0.98,"text":"\"Real-time spatiotemporal stereo matching using the dual-cross-bilateral"},{"category_id":15,"poly":[182.0,1432.0,838.0,1432.0,838.0,1464.0,182.0,1464.0],"score":0.95,"text":"grid,\" in Proceedings of the European Conference on Computer Vision"},{"category_id":15,"poly":[182.0,1458.0,838.0,1458.0,838.0,1490.0,182.0,1490.0],"score":0.98,"text":"(ECCV), Lecture Notes in Computer Science, pp. 510-523, September"},{"category_id":15,"poly":[182.0,1477.0,243.0,1483.0,241.0,1511.0,179.0,1505.0],"score":1.0,"text":"2010."},{"category_id":15,"poly":[134.0,1508.0,836.0,1508.0,836.0,1538.0,134.0,1538.0],"score":0.98,"text":"[20] S. Paris and F. Durand, \u201cA fast approximation of the bilateral filter using"},{"category_id":15,"poly":[182.0,1533.0,836.0,1533.0,836.0,1565.0,182.0,1565.0],"score":0.98,"text":"a signal processing approach,\u201d Int. J. Comput. Vision, vol. 81, pp. 24-52,"},{"category_id":15,"poly":[185.0,1561.0,282.0,1561.0,282.0,1586.0,185.0,1586.0],"score":0.98,"text":"Jan. 2009."},{"category_id":15,"poly":[134.0,1584.0,836.0,1584.0,836.0,1613.0,134.0,1613.0],"score":0.98,"text":"[21] Q. Yang, L. Wang, R. Yang, S. Wang, M. Liao, and D. Nist\u00e9r, \u201cReal-"},{"category_id":15,"poly":[182.0,1609.0,838.0,1609.0,838.0,1641.0,182.0,1641.0],"score":0.98,"text":"time global stereo matching using hierarchical belief propagation.\u201d in"},{"category_id":15,"poly":[182.0,1634.0,698.0,1634.0,698.0,1666.0,182.0,1666.0],"score":1.0,"text":"British Machine Vision Conference, pp. 989-998, 2006."}],"page_info":{"page_no":5,"height":2200,"width":1700}}]
