Stereo Reconstruction Based on Local Edge Detection and Binocular Stereo Matching

A stereo reconstruction algorithm is proposed, which combines techniques of local edge detection, stereo matching and binocular vision. Firstly, the left and right images of the object are taken by calibrated binocular cameras. Then, in rectified matching images, the closed local edge is formed surrounding each pixel by Prewitt edge operator in order to construct an adaptive window, in which cost aggregation is calculated by normalized cross-correlation-coefficient. Next, local winner-take-all optimization is used to obtain disparity, and image smoothing technique is introduced to reduce matching errors. Finally, points cloud of the object is produced based on calibration information of binocular cameras and results of stereo matching. The experimental results show that the proposed algorithm can overcome impacts of real environment on reconstruction system and successfully achieves smooth and vivid 3D points cloud models of different real objects with high automation degree.