The Design And Application Of A Computer-based Binocular Vision System

Binocular stereo vision technology is one of the research hotspots in the field of computer vision. Stereo vision is applied in many fields such as robot navigation, object recognition, virtual reality, aviation, industrial measurement, gesture detection and controlling of micro-operation system, etc.. A computer-based binocular stereo vision system normally includes the following tasks:image acquisition, camera calibration, features extraction, stereo matching, depth restoration and post-processing. Among these tasks, stereo matching to an acceptable accuracy for all situations is one of the key tasks.The paper introduces a convergent binocular stereo vision system, with an emphasis on camera calibration, features extraction, stereo matching and depth restoration. For camera calibration, Harris corners are used as the features. The intrinsic and extrinsic parameters of two cameras are obtained using calibration method of Zhang. Due to the obvious differences between the pictures taken by the left and right cameras of the binocular stereo system, object matching based on image local invariant features are adopted. The Harris corner and SIFT feature are chosen as the matching feature separately. NCC and Euclidean distance are used for the calculation of similarities of two kinds features, respectivly. The matched points are filtered further using RANSAC algorithm. According to the camera imaging principle, over-determined equations which describe the relationship between correspondences’coordinates in image coordinate system and the corresponding object point’s coordinate in world coordinate system are got. Then the distance and the target can be detected by using least square method to solve equations.The binocular stereo vision system is realized using Visual C++and OpenCV. A few sets of images have been tested with the system and acquired good results..