Three-dimensional reconstruction method based on jitter optimization

Abstract

For obtaining richer visual information and eliminating mismatched points faster, a jitter-optimized 3D reconstruction algorithm is proposed. First, we built a stereo vision system that can capture images from multiple views of a jittering plane mirror. A vector-based method is defined to evaluate the performance of 3D point cloud and predict the optimal jittering state of the object, and then select robust feature points with integration optimization strategy. Experimental results show that the proposed algorithm can efficiently eliminate mismatched points and supplement robust features. Furthermore, selecting a reasonable visual angle of the binocular cameras will significantly improve the effect of jitter optimization, and higher-quality reconstruction will be obtained. The proposed method can solve the difficult problem to change the pose of the object or adjust the position of the binocular cameras to capture images from multiple views.