M. Kardouchi & E. Hervet Université de Moncton Moncton, NB, Canada 30-mai-03 1
� Introduction � Approach Feature points extraction points extraction and matching and matching Feature Dense Dense vector field vector field Delaunay triangulation Delaunay triangulation Adaptive interpolation interpolation Adaptive � Parallelization � Results and Conclusion 30-mai-03 2
� Disparities between stereoscopic Disparities between stereoscopic images images � → Displacement between 2 image points � Estimation of dense Estimation of dense correspondences correspondences � → Block matching, Optical flow methods Pb: disparity discontinuities → Feature points matching and Delaunay triangulation 30-mai-03 3
� Extraction of feature points with Harris Detector Left and right images Extracted feature points Harris detector’s threshold : 800-1200 corner points from a 720 x 480-pixel image 30-mai-03 4
� → Maximizing a cross-correlation measure 600-900 acceptable matches for 800-1200 corners points Correspondence vectors between left and right images Problem : small number of mismatches 30-mai-03 5
� Eliminating mismatches → Three steps 1. Delaunay triangulation 30-mai-03 6
2. Identify neighbours of each correspondence point p ( ) η Neighborhood of correspondence point p p 30-mai-03 7
3. A correspondence point p is considered unreliable if: − > ρ d d p p disparity vector at p d p ( ) η average correspondence vector over p d p ρ adjustable threshold. 30-mai-03 8
Correspondences vectors before and after post-processing ~ 16% of correspondences removed 30-mai-03 9
� Triangulation Triangulation and correspondence vectors and correspondence vectors: dense : dense vector field vector field � Assumption: in spatial : in spatial coordinates coordinates, , disparity disparity varies varies linearly linearly Assumption within each triangle. triangle. within each q d q d x p d r p d r x Triangulation Disparity vector ( ) = − − + + d 1 i j d id jd x p q r 30-mai-03 10
Disparity field obtained with linear interpolation Advantage: continuous dense disparity field Limit: disparity discontinuities 30-mai-03 11
Adaptive interpolation - Euclidian distance disparity / 3 vertices - Weight of vertices according to prediction error Disparity field obtained with adaptive interpolation 30-mai-03 12
• To speed up the adaptive interpolation → video flows processing in real time → object tracking • Theoretically a simple linear system 1 30-mai-03 13
• In practice: computation balancing of triangles • First sort triangles according to their number of points • Then processor #1 is given t or t+1 triangles to process: 1 • Repeat until all triangles are processed 30-mai-03 14
• Cluster: 1 server + 16 nodes - server : Intel PIV processor 2GHz, 512 Mb RAM - nodes : Intel PIII 500 MHz, 128 Mb RAM 1 - Ethernet cards 100 Mbits/s - Operating System: Linux Mandrake - Parallel development environment : MPI ( Message Passing Interface ) 30-mai-03 15
1 node: 414 ms 4 proc.: gain 3.45 1 → Standard stereoscopic video: 1 image out of 3 8 proc.: gain 6.18 → video: 1 image out of 2 30-mai-03 16
� Fast method to estimate large disparities � Takes into account disparity discontinuities � Very efficient computationally � Future applications: – Real-time object tracking – Real-time 3D reconstruction 30-mai-03 17
Recommend
More recommend