Applied Research On Double Webcam Stereo Vision System

Binocular stereo vision system is one of the most important researches of computer vision field. With the universal using of webcams, constructing a binocular stereo vision system with two webcams become convenient, on which a man-computer interaction system can be constructed base. The system can be used for operating computers and playing interaction games and so on. Researches on this field will bring people unusual experiences when they using computers. It will change people's lives and has a bright future. So it is a research of theoretical interest and also of great relevance to applications.When a binocular stereo vision system is achieved, two digital images of the target object will be captured by two cameras from different positions. The 3D coordinates of the target object will be got while the parameters of two cameras have known, based on the optical parallax theory. There are four parts in a binocular vision system, such as camera calibration, feature point detection, stereo correspondence and 3D reconstruction. This thesis concentrates on these aspects, analyzes and summarizes the domestic and international researchers' achievements about stereo vision. Through theoretical derivations and experiments considering the needs of applications several progresses have been got as follows: A calibration method which fits for CMOS webcams has been proposed after comparing the features of CMOS webcam and CCD camera. This method has improved Tsai's two-stage method and improves the accuracy of calibration. A HSV model based method which has been used for discerning feature objects and capturing the coordinate of their center using H-parameter Eigen value has been achieved. This method not only detects feature points accurately but also makes stereo correspondence easier. Double Webcam Stereo Vision System has been applied in Development Engine Technology Platform of Interactive Animation Game successfully.The experiment results show that camera calibration method, feature point detection method, stereo correspondence method and 3D reconstruction method used in this thesis fulfill the need of real application completely.