Research On Multi-core Implementation Method Of 3D Object Reconstruction

Parallel 3D object reconstruction in the multi-view environment is the focus of computer science research and has important applications. However 3D object reconstruction is compute-intense and has huge data, processing time is one constraint of 3D object reconstruction applications. The speed of 3D object reconstruction is required as fast as possible in many 3D applications. The acceleration of 3D object reconstruction speed has important theoretical and practical significance. The parallelism of 3D object reconstruction algorithm is explored first, and then the 3D object reconstruction algorithm is parallelized to accelerate the speed of 3D object reconstruction.The main works and features are as follows:1) In order to accelerate the speed of point cloud reconstruction, a parallel method is presented to reconstruct point cloud in the multi-view environment. First of all, the key views were selected by the texture feature discrepancy which is computed based on the grey level grows matrix. Then, Harris corner detection algorithm and Blob detection algorithm are adopted to extract feature of the key view images. Finally, the method is parallelized by exploring the inherent parallelism of proposed procedures. Lastly, a parallel point cloud generation method is presented to accelerate the speed of point cloud generation.2) In order to accelerate the speed of voxelization, this process is parallelized with data decomposition method. Voxelization is a graph cut problem to label the voxels with global energy minimization, and labels the voxels as foreground voxels or background voxels. The bounding box is divided into uniform data blocks, and the multi-core parallel label the voxels in the blocks.3) In order to accelerate the speed of extracting the surface with higher-order smoothness, a parallel surface extraction method is presented. Surface Extraction from Binary Volumes with Higher-Order Smoothness (SEBVHOS) is an approach to extract surfaces in 3D reconstruction. Firstly, SEBVHOS algorithm is parallelized by exploring its inherent parallelism. Secondly, several performance optimization techniques are applied to improve the performance. Meanwhile, matrix compression is applied to improve the performance of memory space.4) In order to accelerate the color recovery speed of the 3D object surface, the color recovery algorithms are parallelized. The surface is in the form of triangle mesh, the vertices of triangle mesh are assigned color to recover the surface color. The color recovery needs to determine the relationship between vertices and views. The averaging method and the best-view method are parallelized to parallel assign color for the vertices.