Study And Implementation Of 3D Scene Reconstruction

This thesis has studied the theory of three-dimensional reconstruction based on two views, and designed a 3D reconstruction software which has been applied in our project.At the beginning of this thesis, we described the development in the field of computer vision, as well as the basic concepts and applications of three-dimensional reconstruction. Then for the three-dimensional reconstruction based on two views, we gave a detailed analysis of each step. Firstly, we compared Harris, SIFT and KLT algorithms, analyzed the experimental data, and proposed a method to be applied in 3D construction. Secondly, we introduced camera imaging models and camera calibration methods, gave the results of the camera calibration. Thirdly, we presented imaging model of the binocular vision system and epipolar geometry. Next, we obtained more accurate matching points according to epipolar geometry constraints, as well as the gray cross-correlation constraints and affine transformation constraints. Finally we calculated the spatial coordinates of matching points with the principle of triangulation.In this thesis, we designed the software of 3D reconstruction, pointed out the problems in the current implementation, and gave suggestions for further improvements.