Research On 3-D Reconstruction Based On Stereo Camera Calibration

Acquiring object's 3-D geometric information from multiple images is one important research topic in the domain of computer vision. The outline of objects can be measured by observing from two different positions, which is generally called 3-D shape measurement. It is an important method for getting 3-D information. The advantage of the method is that it doesn't need special devices. With the development of science, technology and industrial production, 3-D shape measurement has become more and more important in modern industry and practical production. It finds extensive applications in science research, engineering design, trail analysis on the criminal investigation spot, online detection automatically, quality control, machine vision, medicine diagnosis and etc.This thesis analyzes and summarizes the domestic and international researchers'work about stereo vision, and then concentrates on the research on some key techniques of stereo vision, such as camera calibration, corner detection, 3-D matching and 3-D reconstruction. The main progress in this thesis is as follows:1. An improved method for camera calibration is presented. Based on Mr. Zhang zhengyou's plane template two-step method, three aspects of improvements have been made. Firstly, tangential distortion adjustment is added into the camera imaging model. Secondly, after the detailed research aiming at the step of computing the initial guess of camera parameters, some improvements are made within the corner selection and computation. When the distortion effect is not considered, not all the points are selected, but those around the center of image whose distortions are relatively smaller. At last, using the orthonormal property of the rotation matrix and the relation among the parameters from transformation matrices, a comparatively more simple method is used to decompose and derive the camera parameters from the projection matrix.2. A policy that that makes it more accurate step by step is adopted for corner detection, which effectively solve the problem that the illusive edge points caused by noise can't be distinguished using the conventional edge extraction algorithm. Hough transform is firstly executed to deserve the region where the corner locates. Then, Harris operator is used to estimate the rough location and finally gradually acquire the precise location.3. A stereo matching algorithm based on ordering coherence constraint is presented. The difficulty existing in stereo matching is how to eliminate the mismatching, and the effective means is reducing the search region. Firstly, matching feature points in the image are determined. Then, based on the feature points, the...