Research On 3D Reconstruction Based On The Binocular Stereo Vision

Stereo Vision is the classical problem in Computer Vision Area. Binocular Stereo Vision becomes the famous research in Computer Vision Area, due to its eye-similar characteristic. This paper introduces the research content of the Computer Vision and the general situation of its development, and then based on the 3D reconstruction of the terrain data, this paper includes the research on relative orientation, stereo matching and 3D reconstruction.In the stage of relative orientation, we use a man-machine interaction method to select couples of control points from the stereo images, and then calculate the parameters between the stereo camera coordinates and the reference coordinate. After that, we use a new method of projection and resample to realize the calibration of the stereo images and make sure that the stereo images are restrict to the rule that the epipolar lines are parallel in ideal Binocular Stereo Vision system.In the stage of stereo matching, in order to get dense disparity data, we use the area matching method and judge by the maximum correlative coefficient rule. Meanwhile, we compare the effect caused by different window size in area matching. In the strategy of stereo matching, we use a matching method that based on multi-grid to improve the computation efficiency.In the stage of 3D reconstruction, we present a new method of calculating the 3D coordinate of the target point by analyzing and comparing several common methods in 3D reconstruction, and we design experiment to test the algorithm, the result shows that this new method has a higher precision compared to the other algorithms that mentioned in this paper. Also, we use OpenGL as visualize tool to display the 3D reconstruction result, to make the 3D model looks more visual.At last, we do experiment through the procedure of stereo image reading, relative orientation, image stereo matching, disparity data processing and 3D reconstruction, realize the 3D reconstruction system based on Visual C++6.0 and OpenGL under windows environment.