A novel rapid point-cloud surface reconstruction algorithm for laser imaging radar

Abstract

In order to obtain the fast three-dimensional surface reconstruction from given scattered point clouds, a novel improved point-cloud surface reconstruction algorithm for laser imaging radar is proposed so as to reconstruct the three-dimensional depth surface from the depth data and image data in this paper. Firstly, the three-dimensional space is partitioned into voxels with local distance points and finds outliers with point histogram features; then the Gaussian process (GP) regression is adopted to generate a plane similar to a Gaussian distribution; finally, the high-resolution gray data and three-dimensional interpolation points are fused by using Markov random fields to build a dense three-dimensional depth surface. Experimental results show that our proposed algorithm will greatly improve the robustness and reconstruction accuracy of three-dimensional surface reconstruction algorithm and can be used to assist unmanned driving in complex urban scenes.