The Investigation Of 3D Scene Reconstruction Techniques Based On Laser Scan Data

3D modeling technology for reconstruction of real scenes based on range data acquired by laser scanners has broad applications, such as digitizing historic sites and digital amusement. Recently, with decrease in the cost of scan laser and development in laser scanning techniques, the interest in the 3D modeling of real scenes from range data increases. However, the high complexity of the real world results in many difficulties and challenges. Combining the 3D scene reconstruction project cooperated with Dalian Jingtuo Company, the dissertation pay more attention to the relative localization, global localization of the 3D scene reconstruction system, and the surface modeling techniques in 3D scene reconstruction.In the field of relative localization techniques, lots of researches and comparisons are done among various relative localization techniques of mobile system. We focus on the scan matching based relative localization algorithm ICP (Iterative Closest Point). With improvements in point selection and pose estimation, a effective improved ICP-based scan matching algorithm is proposed. The algorithm reduces the calculating time while remains the localization accuracy, and it is especially available for the 3D scene reconstruction with large number of sampling points.In the field of global localization techniques, priori map based and GPS based global localization techniques are analyzed. The Monte-Carlo-Localization (MCL) algorithm in the field of priori map based global localization techniques is discussed detailedly. And a 3D MCL algorithm is proposed in order to rebuild the 3D localization of the reconstruction system by extending the 2D MCL algorithm. Besides, a global localization solution in large scale 3D scene reconstruction is proposed by combing the scan matching and GPS base localization techniques.The main techniques during surface modeling in 3D scene reconstruction are discussed in great detail. With data registration, filter, interpolation, smoothing, mesh generation and texture mapping, surface model is rebuilt. And during texture mapping, an automatic seamless image mosaic method based on feature points is proposed, which is suitable for seamless image mosaic in different view points with various illumination and scale.