3D Point Cloud Reconstruction Based On Image Sequences

Three-dimensional point cloud reconstruction based on image sequence is the processure of using methods in computer vision to deal with the information in the image,so as to obtain the three-dimensional coordinate value of the corresponding object point in the scene.In view of this,it can also be understood as the inverse process of camera shooting process.The whole processure mainly includes the research of three subproblems which are the reconstruction under dual-view situation,the calculation of the transfer parameters between adjacent coordinate system of dual-view and the reconstruction under multi-view situation.The main contents of the work includes:Reconstruction under dual-view situation.In the dual-view reconstruction phase,we use the traditional binocular vision reconstruction methods to complete the processure of camera calibration,image registration and the calculation of point cloud in turn.In view of the problem that the obtained point cloud restricted from the quantity and quality of the matches in the traditional reconstruction methods.A region-growing approach was proposed to get the dense matching results.At the same time,in the processure of the feature matching,the ratio constraint,the symmetry constraint and the epipolar constraint were introduced.During the region-growing processure the disparity gradient constraint was also introduced to improve the accuracy of image registration.After filtering,expanding and refiltering,both the quality and quantity of the matches were optimized.Computation of the transfer parameters between adjacent pairs of dual-view.All these transformation arguments including the rotation matrix,the translation vector and the scale factor of the coordinate system of the reconstructed point cloud under adjacent pairs of dual-view were computed using the characteristics of the continuity of the camera motion.Reconstruction under multi-view situation.Since the traditional reconstruction method using binocular vision strategy cannot get the closed 3D model.Through combining the point cloud and the transformation arguments together,all these point cloud under dual-view can be integrated into the same global coordinate system,and then the reconstruction under multi-view situation were completed.Finally we optimized the reconstruction result,and the result shows that much more dense,complete and accurate 3D model was achieved using the proposed method.