Abstract:

Based on camera selfcalibration of image sequence, we propose an improved bidirectional SIFT feature matching algorithm to solve the problems of high mismatching rate and low operation efficiency of the SIFT algorithm. To remove the mismatching, two steps are to follow. First, the SIFT bidirectional matching algorithm is leveraged to eliminate part of the mismatching. Second, disparity gradient constraint and the RANSAC algorithm are used to purify the matching points. Regarding the speed improvement, we utilize the K nearest algorithm at the beginning of matching, and disparity gradient constraints are also adjusted, reducing the iteration  count to lower the time consumption. Experimental results show that the proposed algorithm removes most of the mismatching and retains enough matching points to automatically calibrate the position and orientation of the cameras in the space, and it can effectively reduce the execution time.

Key words: SIFT;bidirectional matching;disparity gradient;RANSAC algorithm