Scale invariant control points based stereo matching for dynamic programming
Kunpeng Li, Sun’an Wang, Mingxin Yuan, Naijian Chen · 2009
A stereo matching algorithm based on scale invariant control points is proposed for dynamic programming. Firstly, for the problem of large amount of calculation on SIFT feature extraction algorithm, the stable feature points are extracted from left and right image respectively and described by improved SIFT algorithm. Secondly, basing on the description of feature points, the kd-tree nearest neighbor search algorithm is adopted to match feature points, and then these matched feature points are looked as true matching points. Finally, the whole matching process is finished by using the true matching points as the control points. The experimental results show the method can alleviate the effect of horizontal streaks caused by traditional dynamic programming algorithm and the experimental matching results are satisfactory.