Three Dimensional Reconstruction And Obstacle Detection Based On Stereo Vision

Stereo vision is the simulation of biological visual system,using two cameras from different angles or different time to catch two digital images for the same scene,calculate the position deviation of pixels in two image pixels through the stereo matching algorithm to get the 3D geometric information and depth information to reconstruct 3D shape and position of the scene.The process of stereo vision includes camera calibration,distortion correction,stereo matching and so on.This paper is based on this process to complete the following research contents:Using Zhang Zhengyou chessboard method to finish the parameter calibration and distortion correction of the experimental camera,while getting the conversion relationship between the photographing picture and the world coordinates.At the same time,the work of image denoising and region segmentation is completed.Learning three kinds of stereo matching algorithm,use standard test image to test the algorithm performance from the matching time and dense disparity map.Repair the disparity map by interpolation,proposes a improved algorithm for disparity repair.According to the characteristics of disparity map,divide the photographing environment into different regions,finding obstacles of rough position by contour extraction and area feature selection.Aiming at the ground area interference caused by linear distortion,this paper propose a new method based on geometry information to remove the interference and get the distance and cross section size of obstacles.At the same time,image stitching and 3D reconstruction are carried out.The 3D image is displayed by OpenGL,and the 3D point cloud is retained.Building the hardware system of visual experiment,developing vision measurement software,the error data of obstacle distance measurement were analyzed and summarized,and the applicability of the system was summarized.