Stereo Real-time stereo on GPU Stereo Challenges

Stereo Real-time stereo on GPU Stereo Challenges www.phwiki.com

Stereo Real-time stereo on GPU Stereo Challenges

Taylor, Rebecca, Morning Anchor has reference to this Academic Journal, PHwiki organized this Journal Stereo Class 7 Read Chapter 7 of tutorial http://cat.middlebury.edu/stereo/ Tsukuba dataset Geometric Computer Vision course schedule (tentative) Stereo St in addition to ard stereo geometry Stereo matching Aggregation Optimization (1D, 2D) General camera configuration Rectifications Plane-sweep Multi-view stereo

Eugene Bible College US www.phwiki.com

This Particular University is Related to this Particular Journal

Real-time stereo on GPU Computes Sum-of-Square-Differences (use pixelshader) Hardware mip-map generation as long as aggregation over window Trade-off between small in addition to large support window (Yang in addition to Pollefeys, CVPR2003) 290M disparity hypothesis/sec (Radeon9800pro) e.g. 512x512x36disparities at 30Hz GPU is great as long as vision too! Stereo Challenges Ill-posed inverse problem Recover 3-D structure from 2-D in as long as mation Difficulties Uni as long as m regions Half-occluded pixels

Pixel Dissimilarity Absolute difference of intensities c=I1(x,y)- I2(x-d,y) Interval matching [Birchfield 98] Considers sensor integration Represents pixels as intervals Alternative Dissimilarity Measures Rank in addition to Census trans as long as ms [Zabih ECCV94] Rank trans as long as m: Define window containing R pixels around each pixel Count the number of pixels with lower intensities than center pixel in the window Replace intensity with rank (0 R-1) Compute SAD on rank-trans as long as med images Census trans as long as m: Use bit string, defined by neighbors, instead of scalar rank Robust against illumination changes Rank in addition to Census Trans as long as m Results Noise free, r in addition to om dot stereograms Different gain in addition to bias

Systematic Errors of Area-based Stereo Ambiguous matches in textureless regions Surface over-extension [Okutomi IJCV02] Surface Over-extension Expected value of E[(x-y)2] as long as x in left in addition to y in right image is: Case A: F2+ B2+(F- B)2 as long as w/2- pixels in each row Case B: 2 B2 as long as w/2+ pixels in each row Right image Left image Disparity of back surface Surface Over-extension Discontinuity perpendicular to epipolar lines Discontinuity parallel to epipolar lines Right image Left image Disparity of back surface

Over-extension in addition to shrinkage Turns out that: as long as discontinuities perpendicular to epipolar lines And: as long as discontinuities parallel to epipolar lines R in addition to om Dot Stereogram Experiments R in addition to om Dot Stereogram Experiments

Offset Windows Discontinuity Detection Use offset windows only where appropriate Bi-modal distribution of SSD Pixel of interest different than mode within window Compact Windows [Veksler CVPR03]: Adapt windows size based on: Average matching error per pixel Variance of matching error Window size (to bias towards larger windows) Pick window that minimizes cost

Integral Image Sum of shaded part Compute an integral image as long as pixel dissimilarity at each possible disparity Results using Compact Windows Rod-shaped filters Instead of square windows aggregate cost in rod-shaped shiftable windows [Kim CVPR05] Search as long as one that minimizes the cost (assume that it is an iso-disparity curve) Typically use 36 orientations

Locally Adaptive Support Apply weights to contributions of neighboring pixels according to similarity in addition to proximity [Yoon CVPR05] Locally Adaptive Support Similarity in CIE Lab color space: Proximity: Euclidean distance Weights: Locally Adaptive Support: Results Locally Adaptive Support

Taylor, Rebecca KVOA-TV Morning Anchor www.phwiki.com

Locally Adaptive Support: Results Occlusions (Slide from Pascal Fua) Exploiting scene constraints

Ordering constraint 1 2 3 4,5 6 1 2,3 4 5 6 2 1 3 4,5 6 1 2,3 4 5 6 surface slice surface as a path occlusion right occlusion left Uniqueness constraint In an image pair each pixel has at most one corresponding pixel In general one corresponding pixel In case of occlusion there is none Disparity constraint surface slice surface as a path bounding box disparity b in addition to constant disparity surfaces

Plane-sweep multi-view matching Simple algorithm as long as multiple cameras no rectification necessary doesn’t deal with occlusions Collins’96; Roy in addition to Cox’98 (GC); Yang et al.’02/’03 (GPU) Next class: structured light in addition to active scanning

Taylor, Rebecca Morning Anchor

Taylor, Rebecca is from United States and they belong to KVOA-TV and they are from  Tucson, United States got related to this Particular Journal. and Taylor, Rebecca deal with the subjects like City/Metropolitan News; International News; Local News; National News; Regional News

Journal Ratings by Eugene Bible College

This Particular Journal got reviewed and rated by Eugene Bible College and short form of this particular Institution is US and gave this Journal an Excellent Rating.