Research Of Several Essential Problems Of Stereo Vision

Stereo vision is a research hotspot of the computer vision domain, it's objective is the restoration of three-dimensional structure of a scene based on geometry principle requiring two or more cameras to observe the identical scenery shown at a few different orientations. Along with the development of computer technology, stereo vision will be widely applied in the production automation, automatic navigation, the medicine, information security as well as antiterrorism domains, etc. Therefore the research of stereo vision has the important theory significance and the practical value.Based on the analysis and research of much technology and literatures concerned, this paper takes the Marr vision theory as the starting point, and aims at the basic problems of stereo visual domain such as the stereo match, the camera calibration and the 3D reconstruction and so on, and uses the method of combination of theoretic analysis and experimental research to carry on the systematic comparison and analysis of the present commonly used methods of stereo match and camera calibration, and proposes some own solutions of the problems which are not yet resolved faultlessly. In this paper, the main innovative ideas are as following:1. A novel feature point matching algorithm which is based on adaptive selection of disparity is presented aimed at solving the problem of most feature point matching algorithms' poor matching accuracy and computational efficiency in the presence of high proportion outliers. Practical calculation result shows that the new algorithm can achieve not only high accuracy, but also high computational efficiency matching result, thereinto, 97.6% of outliers can be correctly classified, which is respectively a increase of 59 and 7 percentage points with respect to current LMedS(Least-Median-Squares) method and the latest proposed ASSC (Adaptive Scale Sample Consensus) method;The computational time is only 0.078 seconds which is respectively a reduction of 5 and 6 orders of magnitude with respect to LMedS and ASSC. 2. Based on the research objective of this paper and the equipment condition of our laboratory, author implements three commonly used planar-pattern-based camera calibration Techniques which proposed by Zhang, Bouguet and Heikkila, and presents a comparative analysis of the three techniques in terms of accuracy, speed, application and interrelation to supply references to succedent studies.