A Research And Improvement Of The Binocular Vision Stereo Matching Algorithm Based On Adaptive Support Weight

Binocular vision stereo matching is an important branch of computer vision, aimed at finding an effective method to extract three-dimensional information from two-dimensional images, restores the three-dimensional scene. Currently stereo matching algorithm is divided into two categories: global algorithm and local algorithm, global algorithm owed high precision, but a very high computational complexity; although local algorithm computing faster, but matching accuracy was worse than the global algorithm. Yoon proposed an area matching algorithm based on adaptive support weight that the match accuracy of local matching algorithm has reached a very high level. But there are some problems in this algorithm, such as the calculation of the weights aggregation is in a high computational complexity; in addition, the post-processing of this algorithm is also an important issue.After a lot of research on the local matching algorithm, this paper proposed an improving algorithm based on adaptive support weights algorithm, which has been improved the performance of the old algorithm. The main research work and innovation of this paper is as follows:1. Briefly analyzed the stereo matching model and the main currently stereo matching algorithm, introduced the general process of area matching and the evaluation criteria of stereo matching. Deeply researched the principle of adaptive support weight algorithm and analyzed four corrective algorithm based on adaptive support weight algorithm. Through experiments we obtained four correction algorithm’s performance and computation speed, then chose the two-pass adaptive support weight method in improving algorithm.2. Researched deeply in the disparity refinement method for area matching in post-processing step, proving the effectiveness of left-right consistency through experiment, in order to illustrate the importance of the disparity refinement step. Introduced an iterative disparity refinement method to deal with the initial disparity map, and detailed analyzed the principle of iterative disparity refinement method, then made several experiments, analyzed the results, select the number of iterations in this article to the improving algorithm is8times.3. Proposed an improving algorithm based on adaptive support weight algorithm, improved the algorithm matching’s performance in matching accuracy and computation speed. First, we use two pass ASW instead of ASW algorithm in the step of cost aggregation, effectively reducing the computation time; then use the iterative disparity refinement method in post-processing, improved matching accuracy. Then made experiments on Vision Studio2010platform, proved that the improving algorithm’s efficacy and superiority compared with the old algorithm on matching accuracy and computation time through experimental results.