The Research Of Stereo Matching Algorithm Based On Binocular Vision

Binocular vision technology is widely used in object recognition, virtual reality, industrial inspection, robot navigation and other aerospace fields. Binocular vision includes image acquisition, camera calibration, dimensional calibration, stereo matching and three-dimensional information reconstruction. Stereo matching is an important research topic in stereo vision.This paper analyzes the basic theory of binocular stereo vision and parallax theory. Summarize and compare the advantages and disadvantages of the three types of stereo matching algorithms, which are based on local, global and semi-global. Describe the evaluation of stereo matching algorithm. The minimum spanning tree stereo matching algorithm has been studied deeply and improved. The main work is as follows:The minimum spanning tree stereo matching algorithm only takes into account one channel of the R, G, B channel, ignores the effects of the other channel to the final edges wight in color images. This paper proposes a combined three-channel edge weights, which calculate the weighted euclidean distance using a three-channel approach, and this algorithm has been tested on Tsukuba、Venus、Teddy、Cones standard image. Simulation results show that the improved algorithm ameliorated the robustness of the edge weight function and the connectivity of the minimum spanning tree, not only improved matching accuracy, but also the computational speed is faster than the original algorithm.An experimental platform is set up for health care, and the improved matching algorithms are tested. Designed camera calibration, epipolar rectification, stereo matching experiments and validate the usefulness of the algorithm. Use the camera calibration parameters and the disparity map reconstructed point cloud image of the target person, from which can get character height, distance and three dimensional coordinates of the camera.