The Parallelization Of Research For Point Cloud Reconstruction Based On Heterogeneous Environment

With the rapid development of computer vision and computer graphics technology is widely used in various fields, as well as intelligent simulation technology, the relevant technology of3D reconstruction has been researched as focus by many research scholars. The process of point cloud reconstruction use the feature point information extracted from multi-view, to match feature points and generate patches by extend, which is the basic process of3D reconstruction based on multi-view.The Patched-based Multi-View Stereo(PMVS) algorithm, and the parallel technology based on multi-core CPU or GPU architecture are introduced after the basic process of point cloud reconstruction. PMVS algorithm mainly extracted feature points extracted by Harris and DoG algorithm, after complete feature matching, extended and filter to generate a3D model. Since including the whole process of the point cloud reconstruction, so the parallel study is aimed at the point cloud reconstruction in PMVS algorithms. Through parallel analysis and serial experiments for point cloud reconstruction, found that the date of feature matching and extended is independent, and the calculation is largely. Therefore, parallel study aimed at these two processes is based on compilation model about multi-core CPU, GPU architecture, and CPU/GPU hybrid parallel heterogeneous environment. And do parallel experiments on the different scale of kermit and hallFeng data sets, which on different experimental platform respectively. Finally, analysis the efficiency, accuracy and scalability about the result of reconstruction, shows that the process of point clouds reconstruction have parallelism and scalability.