车辆曲面重构中点云精简算法的研究与改进

发布时间：2018-10-31 20:20

【摘要】：为了解决车辆点云数据曲面重构效率低和精简后数据重构模型质量差的问题,提出一种改进的点云精简算法。基于kd-tree建立散乱点云数据的空间索引结构并获取每个数据点的k邻域索引;提出基于快速识别边界线的精简算法避免精简过程边界数据丢失,确保获得真实的车辆曲面重构模型;对非边界点邻域进行区域分类,并根据分类选择性保留邻域数据,以提高点云数据处理速度并减少内存开销。在实现了算法的程序设计及仿真实验的基础上,完成了基于三维激光扫描车辆外廓尺寸测量系统平台的实车实验。实验结果表明,改进后的精简算法程序最大限度地保留了车辆点云的的边界特征和细节形状,改善了车辆点云曲面重构模型质量;数据处理中能够精简45%~70%的车辆点云数据,加快了系统重构的速度,提高了车辆外廓测量的性能。
[Abstract]:In order to solve the problems of low efficiency of surface reconstruction of vehicle point cloud data and poor quality of data reconstruction model after simplification, an improved point cloud reduction algorithm is proposed. Based on kd-tree, the spatial index structure of scattered point cloud data is established and the k-neighborhood index of each data point is obtained. A simplified algorithm based on fast recognition of boundary line is proposed to avoid the loss of boundary data in the process of simplification and to ensure that the real vehicle surface reconstruction model can be obtained. In order to improve the processing speed of the point cloud data and reduce the memory overhead, the region classification is carried out on the non-boundary point neighborhood, and the neighborhood data is selectively retained according to the classification. On the basis of realizing the program design and simulation experiment of the algorithm, a real vehicle experiment based on 3D laser scanning vehicle profile measurement system platform is completed. The experimental results show that the improved algorithm keeps the boundary feature and detail shape of the vehicle point cloud to the maximum extent and improves the quality of the vehicle point cloud surface reconstruction model. In the data processing, 4550% of the vehicle point cloud data can be simplified, the speed of system reconstruction is accelerated, and the performance of vehicle profile measurement is improved.
【作者单位】：合肥工业大学计算机与信息学院;合肥工业大学汽车研究院;
【基金】：国家重点研发计划项目(JZ2016ZDYF1065)资助
【分类号】：TN249;TP301.6

【相似文献】