Research On Binocular Stereo Vision Image Matching

The technology of binocular stereo vision recovers mainly the interested three-dimensional information from two-dimensional information. This technology is applied to every domain, such as data information fusion, navigation, Reproduce three-dimensional and target identification, etc. The core part of binocular stereo vision is binocular matching,the algorithm of which can directly impact on the effect of binocular vision. Good match algorithm must possess the property of good robustness, fast calculation and High accuracy.Because of high dependence on images and as the most difficult problems in binocular development, match algorithm become the focus of scholars research at home and abroad.First, the discovery process and current situation of binocular stereo algorithm is described. The investigation is in process for existing problems. Binocular stereo vision principle is introduced by bionic perspective. The significance of matching algorithm is analyzed. Then, fundamental theory binocular stereo algorithm is narrated concisely,including optical parallax, matching conversion, matching basic boundary condition, etc.According to function range and element of matching , matching algorithm is categorized.Main and common type is Area matching, global stereo matching and Feature Point matching.Area matching is at the cost of SAD、SSD、NCC operators. Feature Point matching is at the cost of Moravec、SUSAN、Harris operators. The comparison between stereo matching algorithm based on the traditional theory of dynamic programming and stereo matching algorithm based on the traditional theory of Simulated Annealing is made. Pros and cons and application scope of algorithm is investigated..Second, in the paper , the CENSUS,the recently more popular matching algorithm, is regarded as the key research object base on in-depth investigation of local binocular stereo matching algorithm. It is compared with this case,area matching is at the cost of SAD、SSD、NCC operators. For the defect, CENSUS algorithm relies on window center pixel intensity too much, the improvement is proposed by continuity constraint in matching constraint condition the anti-noise character of algorithm is been further enhanced.Third, global matching which is popular matching algorithm is investigated. Aim at heavy computation, long time and poor instantaneity in global matching ,the improvement is proposed base on in-depth investigation of global matching algorithm. Then shortcomings are improved, including reducing time, increasing computation speed and instantaneity.