Key Technologies For 3D Model Reconstruction In Laser Radar Point-Clouds

Laser radar(LiDAR) is an advanced sensor for fast and accurate three dimensional(3D) shape measurement. Point-clouds acquired with a laser radar provide the geometric shape information of a scene. 3D model reconstruction from large-scale point-clouds is a hot research topic in the area of intelligent 3D data processing. It has an increasing number of applications in digital earth, digital city, industrial manufacture, inverse engineering, intelligent transportation, cultural reservation, measurement and remote sensing, anti-terrorism, and criminal investigation. Currently, the challenges for 3D modeling are point-cloud acquisition, registration, mesh reconstruction and texture mapping. The thesis therefore, mainly focuses on point-cloud registration, mesh reconstruction, and texture mapping.First, the point-cloud registration algorithm has been investigated. In order to cope with the high computational complexity of the Coherent Point Drift(CPD) algorithm when dealing with large-scale and high-dimensional poinclouds, a Fast Coherent Point Drift(F-CFD) based point-cloud registration algorithm is proposed. The proposed algorithm adopts a global convergent SQUARed iterative Expectation Maximum(gSQUAREM) technique rather than the Expectation Maximum(EM) technique. Experimental results show that the registration time is significantly reduced while maintaining a high registration accuracy by using the proposed algorithm.Second, the 3D model reconstruction algorithm has been studied. The basic concepts and theories for mesh triangulation are initially introduced. Several classic algorithms including the quad-tree/octree based, Delaunay based, region growing based, and implicit function based algorithms are then presented. The Possion surface reconstruction algorithm and the pivoting ball surface reconstruction algorithm are extensively investigated in the thesis. The principles and characteristics of these two algorithms are reviewed and analyzed. Extensive experiments have been conducted to analyze the reconstruction effectiveness, computational time, and applications domains of these two algorithms.Third, the texture mapping algorithm for 3D models has been investigated. This thesis mainly focuses on the harmonics mapping based constrained texture mapping algorithm. The three modules(i.e., model parametrization, texture coordinate calculation, and mapping optimization) of the harmonics mapping based algorithm are carefully analyzed. Subsequently, a system for texture mapping on typical 3D models and texture images are developed. Experimental results show that the texture mapping algorithm is effective, robust, and suitable for 3D modeling.