Stereo Vision System Based On Real-time Motion Stereo Matching

With 2 or more cameras, stereo vision system can simulate the ability of human eyes system, and calculate the depth information via the differences appears in the image from different cameras. Stereo vision system has shown its wild potential applications in many areas including robot walking navigation, industrial controlling, etc. It is well known that stereo matching with global optimization can obtain good matching result but takes too much computation resources. Real-time stereo system nowadays uses local method in stereo matching, which can run as far as 30 frames per second in a general PC workstation but will a higher miss rate in matching.This paper analyses current real-time stereo vision system and addresses its inadequacy: real-time stereo system nowadays only takes static scene at time t, and ignores all motion features in the moving scene. Recently, many researches shows that with proper use of the motion feature in the scene, stereo matching algorithm can obtain a more reliable result and reduce the miss rate in matching. Based on this, this paper proposes a method that extends existing real-time stereo matching algorithm which makes use of the motion features in the moving scene.It is well known that most structured light strategies are not applicable in moving scene with real-time stereo matching. This paper analyses the reason and proposed a modified strategy based on the M-Array strategy, and then designs an experiment to verify the feasibility of the modified M-Array strategy.In this paper, a stereo vision system based on real-time motion stereo matching is designed and implemented. This system obtains stereo image streams from a pair of calibrated and synchronized cameras. The system can calculate the 3D information of the moving scene in real-time, with more reliable result than other real-time systems.