Based On Binocular Vision Of Three-dimensional Information Acquisition And Reconstruction Method

Human vision is an important means of observing and understanding the world. Scientific research and statistics show that nearly 75% information that human from outside is gained from the vision. Computer Vision is a subject that using computer to simulate human vision in order to perceive and settle the three-dimensional information of the objective world. It gets the image sequences from the camera, then uses computer to realize the image, describes and settles the three-dimensional information of the image by image processing and pattern recognition. Computer Vision's ultimate destination is to observe and realize the world by camera just like what human-beings do. It has a very broad application prospects in computer science and engineering, signal processing, physics, applied mathematics and statistics, neurophysiology, cognitive science, and intelligent application in manufacturing, testing, document analysis, medical diagnosis and military fields.Three-dimensional reconstruction based on binocular stereo vision is a way to reconstruct by computer vision and image processing. Compared with other reconstruction method, this approach doesn't need other aid equipment to help reconstructing, using disparity information which is contained in binocular stereo visual images to reconstruct. Thus this approach is fast efficient and low cost. A complete system of three-dimensional reconstruction based on binocular stereo vision mainly contains binocular stereo vision acquisition platform and software platform. And software platform is divided into five parts which contains camera calibration, feature extraction, epipolar lines rectification, stereo matching and three-dimensional reconstruction.This paper introduced the theories of each parts ordered by the structure of the three-dimensional reconstruction based on binocular stereo vision. The system used two cameras which have different parameters to build binocular stereo vision acquisition platform. Because different parameters will cause a wrong result of the calibration, the paper proposed a method that firstly calibrated each camera separately using Zhang Z Y calibration method and then did stereo calibration to calculate the relative position between two computers in order to reduce the error caused by the different parameters. Before doing matching, the system extracted strong feature points using Canny edges detection. Next the system rectified the two images using the result of the camera calibration so that the epipolar lines are arranged along the image rows in order to improve the speed and accuracy of stereo matching. The system can now do stereo matching with the strong feature points after being rectified. The paper used a method of sparse point matching based on the gray window, which means matching with strong feature points'Neighborhood window's gray-value of the images. After matching all of the feature points, the system will do a linear interpolation to restore the information of the surface. Because there were a small number of feature points and they had strong characteristic, the system can get a reliable result. And this method has an advantage of high speed and precision. Finally the system calculated the three-dimensional coordinate and disparity image using the result of stereo matching and proposed a method of fast three-dimensional Algorithm based on rectified images and did three-dimensional graphic reconstruction with the disparity image using OpenGL API.At last, the paper implemented the algorithm and did lots of experiences with different parameters. The results shows that, in stereo matching part, different window size would cause different matching results and using same window size to match different pairs of images could also cause different results. In other way, when doing three-dimensional reconstruction after been matched, although three-dimensional coordinates of the point may has different result because of the different world coordinate system origin points, the disparity map can basically reflect the distance between the point and camera, and after being graphic reconstruction with OpenGL API, the three-dimensional information of the images is basically resumed.