车辆曲面重构中点云精简算法的研究与改进

Research and improvement of point cloud simplification algorithm in vehicle surface reconstruction

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。

In order to solve the problem that the efficiency of surface reconstruction based on vehicle point cloud data is low and the quality of the reconstruction model based on simplified data is poor, an improved point cloud simplification algorithm is proposed. Firstly, a spatial index structure of the scattered cloud data is established with kd-tree, which obtains the k neighborhood index of each data point. Secondly, a simplification algorithm based on fast identification of boundary lines is proposed to avoid the loss of boundary data in the process of reducing and ensure the real vehicle surface reconstruction model. Finally, the non-boundary point＇ s neighborhood is classified, and the neighborhood is reserved according to the classification, which accelerates the processing speed of point cloud data and reduces memory overhead. The paper not only designs the software program of the simplification algorithm and realizes the simulation experiment, but also carries out a real vehicle experiment on a platform of vehicle-body dimension measurement system based on the 3D laser scanning. The experimental results show that the improved algorithm preserves the boundary features and detail shapes of the vehicle point cloud to the maximum extent, which improves the quality of surface reconstruction. The improved simplification algorithm could reduce the vehicle point cloud data by 45% - 70%, therefore, it improves the speed of surface reconstruction of vehicle point cloud and enhances the performance of the vehicle-body dimension measurement system.