3D Model Reconstruction Based On 2D Images

In this paper, a stereo vision system of 3D model reconstruction from 2D images is designed The system can be divided into five modules, which are image processing,camera calibration,stereo matching,3D point reconstruction and model reconstruction.In the image processing module, the common image processing methods are discussed, and an interactive local corner detection and edge detection algorithms are proposed based on the study of feature points extraction algorithm and edge detection algorithm.In the camera calibration module, projection matrix is corrected by use of nonlinear distortion coefficient obtained from tsai camera calibration, and a new projection matrix in distorted condition is found.In the stereo matching module, a new correlation constraint matching algorithm named RGB correlation constraint is proposed on the base of stereo matching algorithm and matching constraints. Three matching algorithms are proposed for three different types of objects, which are the interactive stereo matching algorithm, feature points matching algorithm and dense matching algorithm.In the 3D point reconstruction module, common 3D points reconstruction algorithms are studied, a detail derivation is given for 3D point reconstruction algorithm based on the midpoint of skew lines common perpendicular algorithm.After getting the calibration parameters of two cameras and completed stereo matched images,3D feature points are reconstructed based on the 3D point reconstruction algorithm.Then the 3D points cloud is filtered, smoothed, triangulized and texured in roder to get a true 3D model.A software platform is developed to integrate all the modules together. we can get an interactive Operating for the key technologies.Some experiments and examples of 3D reconstruction from 2D imaged are given in the thesis, which show the algorithms mention above are pritical and efficient.