Determining point correspondences between two views under geometric constraint and photometric consistency

摘要: Matching or tracking points of interest between several views is one the keystones many computer vision applications, especially when considering structure and motion estimation. The procedure generally consists in independent steps, basically 1) point extraction, 2) matching by keeping only ``best correspondences'' with respect to similarity some local descriptors, 3) correspondence pruning keep those consistent an estimated camera (here, epipolar constraints homography transformation). Each step itself a touchy task which may endanger whole process. In particular, repeated patterns give lots false matches are hardly, if never, recovered 3). Starting from statistical model Moisan Stival, we propose new one-stage approach steps 3), does not need tricky parameters. advantage proposed method its robustness patterns.