复杂曲面结构的激光雷达扫描点云数据处理方法研究

LiDAR (Light Dection And Ranging) scanning is a new kind of high precision and non-contact measurement technology. With 3D LiDAR scanning, dense point cloud of measured object and environment in which the object located can be obtained. By using point cloud processing methods, structure features and location features of the measured object can be acquired from the point cloud. However, by using the existing point cloud processing methods, computation time of procedure of point cloud processing and accuracy of the acquired structure features and location features of the measured objects with complicated curve surfaces are not satisfactory. For the purpose of acquiring more accurate structure and location features of the measured objects with complicated curve surfaces in less computation time, mapping relationship between structure of complicated curve surfaces and distribution of laser points obtained during the scanning process will be studied. Inversion model of curve surface features will be studied as well. Subsequently, a number of novel signal processing methods, such as clustering analysis, wavelet transform and fractional Fourier transform, will be introduced to process the point clouds obtained from the measured objects with complicated curve surfaces by using LiDAR scanning to promote the accuracy of measurement results and computational speed in the procedure of point cloud processing. At last, to validate the novel point cloud processing methods mentioned above, a semi-physical simulation experimentation of 3D LiDAR scanning system will be established. In the experimentation, the point clouds of objects with complicated curve surfaces will be measured effectively. The data obtained from semi-physical simulation experimentation will verify and modify the proposed point cloud processing methods. This work can provide scientific basis for high-speed and high-precision 3D LiDAR scanning system and point cloud processing software independently researched and developed by China.

激光雷达扫描是一种新型高精度非接触式测量手段，可快速获取被测物体及其所处场景的三维点云数据，对点云数据进行处理，可提取出被测物体的表面三维结构及位置信息。利用现有的点云数据处理方法对包含复杂曲面结构被测物体的点云数据进行处理时，存在三维结构和位置信息提取精度不理想及运算速度较慢等问题。为了更快速、准确地获取复杂曲面结构的三维结构及位置信息，本项目将完成以下创新工作。首先，建立复杂曲面结构与激光雷达扫描过程中激光脚点时空分布规律之间的映射关系及曲面特征反演模型；进而，引入聚类分析、小波变换和分数阶傅里叶时频分析等数学方法对复杂曲面结构的点云数据进行分析和处理，以获得更好的信息提取效果；最后，搭建针对复杂曲面结构进行激光雷达扫描的半物理仿真实验系统，对模型和方法的有效性进行验证。本项目研究成果将为我国高精度激光雷达扫描系统及相关数据处理软件的自主研发提供理论依据和验证手段。

激光雷达扫描探测技术作为一种新型高精度非接触式测量手段已经在现代工业制造过程中得到越来越广泛的应用。由于点云数据内激光脚点的分布存在有高冗余性及随机离散性，现有的点云数据处理方法仅可高效地对具有相对较为简单外表面结构的被测物体的点云数据展开分析，进而获取相对理想的检测结果。当被测物体具有较为复杂的曲面外表面结构时，现有方法的分析效果较差。本项目针对这一问题展开研究工作，取得以下成果：(1)针对典型曲面结构，建立其激光脚点特征分布规律与其外表面形态之间的映射关系，从而获得到该曲面结构模型特征的反演方法；(2)将数学形态学处理方法及工业制图中的多视角概念引入曲面点云数据处理过程中，提出一种新型深度影像生成方法，将散乱点云数据处理问题转换为多视角下的三维深度影像处理问题，极大的降低了对被测物体外表面特征信息进行提取的难度；(3)针对特征提取算法展开研究工作，提出一种新的基于特征组合及区别分类器的特征提取方法，利用该方法可有效对高精度深度影像进行处理并从中获取被测物体的特征参数；(4)利用小波变换方法针对曲面被测物体点云数据展开分析，提出一种针对球、椭球、椭圆锥体等二次对称曲面的高精度特征参数提取方法；(5)搭建针对复杂曲面结构进行激光雷达扫描的半物理仿真实验系统，对本项目所研究方法的有效性进行验证，但不仅局限于半物理仿真实验，在港口散杂货垛堆盘点、堆取料机定位、多相流测量中复杂液泡形态测量及基于光电探测技术的液态物体特征探测等方面，应用本项目所获得新技术与新方法进行试验，均取得较好效果。在本项目支持下，已发表及正式录用待发表期刊论文16篇，其中SCI检索论文10篇，EI检索论文5篇；参加国际会议2次，发表EI检索会议论文3篇；申请国家技术发明专利8项，已授权1项；获省部级科研奖励2项。