Research On Stereo Matching Algorithm Of Binocular Vision

Binocular vision technology can restore three-dimension information of a scene by two images taken from different angles. Compared with monocular vision, binocular vision has the advantages such as high accuracy and strong robustness. As an important research direction of computer vision technology, binocular vision has been widely used in many fields such as aircraft or robot navigation, distance measurement, virtual reality, etc. This paper aims at stereo matching algorithm in binocular vision, and improves the algorithm according to a pedestrian counting system.On the basis of summarizing research status of binocular stereo matching and video-based pedestrians counting system, imaging model of cameras, binocular imaging model and constraints are researched in this paper. General steps and common classification of stereo matching is also researched. This paper discusses the advantages and disadvantages of local and global stereo matching algorithms first, then makes a study of region gray scale based stereo matching algorithm and improves it based on pedestrian counting system. To improve the algorithm, box filtering technology is adopted to accelerate the process; crosscheck and similarity filtering are adopted to reduce matching errors. Test results prove that the improved algorithm has good accuracy and real-time, realizing 17 frames per second on DSP system.To improve the performance in extreme light condition, this paper studies stereo matching algorithm based on dynamic programming (DP), which has the best real-time performance among global stereo matching algorithms and makes a optimization on real-time and matching precision. Control point based fast dynamic programming is proposed to improve the rapidity of conventional DP. At the same time, a second time DP is conducted in vertical direction after the horizontal one to alleviate the stripe effect. Test results show that the algorithm has high precision and strong robustness and can meet the accuracy requirement of pedestrian counting system with an average matching error of around 10%.In addition to theoretical research, experiments are conducted in this paper. Pedestrian counting system based on both region-based and global-based matching algorithm is established. Algorithm is improved based on Experiment Result to applies on the pedestrian counting system. Experiment result shows that pedestrian counting system with region-based stereo matching algorithm can obtain an accuracy of more than 95% in the indoor environment, and keep real-time. For the outdoor environment, the system based on improved DP can adapt to extreme illumination conditions.Finally, the content of this paper is summarized, and future work is presented.