A Multi-Window Stereo Vision Algorithm With Improved Performance at Object Borders

Conventional correlation based stereo vision algorithms have poor performance at object borders due to occlusion. In this paper, a new matching sequence has been introduced. To detect left object border, a left shifted window is used with right image as the reference image and process from left to right. This method gives excellent left object border detection while detection of right object border is poor. The right object borders can be detected with equal success using a right shifted window with left image as reference and process from right to left. The combination result is good detection at both object borders