车载激光雷达点云快速栅格化算法研究及应用  

Research and Application of Fast Rasterization Algorithm Based on Vehicle LiDAR Point Cloud

作　　者：李玉萍 潘文武[2] 田俊林[2] 梁文海[1,3] LI Yuping;PAN Wenwu;TIAN J unling;LIANG Wenhai(College of Physics and Electronic Engineering,Sichuan Normal University,Chengdu 610101,China;China Engineering Physics Academy Application Institute of Electronics,Mianyang,621900,China;Key Laboratory of Wireless Sensor Network in Universities of Sichuan Province,Chengdu 610101,China)

机构地区：[1]四川师范大学物理与电子工程学院,四川成都610101 [2]中国工程物理研究院应用电子学研究所,四川绵阳621900 [3]无线传感器网络四川省高校重点实验室,四川成都610101

出　　处：《测绘地理信息》2023年第4期47-50,共4页Journal of Geomatics

基　　金：中国工程物理研究院创新发展基金(C-2020-C2020034)。

摘　　要：针对车载激光雷达点云数据量大、密度高且存在分层错位和噪点等情况,提出了一种具实时性激光点云快速栅格化算法,该算法根据雷达扫描精度预设栅格单元大小,可在不丢失对象形状特征的情况下,能快速完成点云数据平滑及降采样处理,并将数据量缩小为处理前的60%。将该栅格算法处理后的点云数据应用于深度学习,作为pointnet++神经网络的训练集及测试集,完成语义分割模型训练与测试。实验结果表明,该算法可在1 s内完成上百万量级的点云栅格处理,并且经该算法处理后的点云数据能有效缩短训练时长、提升网络测试精度。Aiming at the large amount of data,high density,dislocation and noise in LiDAR point cloud,a real-time laser point cloud rapid rasterization method was put forward.The algorithm presets the size of the grid cells according to the radar scanning accuracy,which can complete point cloud data smoothing and downsampling processing quickly and reduce the amount of data to 60%without losing the shape characteristics of the object.And the rasterized data is applied to deep learning as the training set and test set of the pointnet++neural network to complete the training and testing of the semantic segmentation model.It is tested that the millions of point cloud can be rasterized by rasterization method within 1 second,and the rasterized data can shorten the training time and improve the accuracy of network testing effectively.

关 键 词：车载激光雷达 激光点云 点云栅格化 深度学习 

分 类 号：P234[天文地球—摄影测量与遥感] TP391[天文地球—测绘科学与技术]