动态环境中多帧点云融合算法及三维目标检测算法研究

Research on multi-frame point cloud fusion algorithm and 3D object detection algorithm in dynamic environment

低线束激光雷达扫描的点云数据较为稀疏,导致无人驾驶环境感知系统中三维目标检测效果欠佳,通过多帧点云配准可实现稀疏点云稠密化,但动态环境中的行人与移动车辆会降低激光雷达的定位精度,也会造成融合帧中运动目标上的点云偏移较大。针对上述问题,提出了一种动态环境中多帧点云融合算法,利用该算法在园区道路实况下进行三维目标检测,提高了低线束激光雷达的三维目标检测精度。利用16线和40线激光雷达采集的行驶路况数据进行实验,结果表明该算法能够增强稀疏点云密度,改善低成本激光雷达的环境感知能力。

The point cloud data scanned by the low-beam LiDAR is relatively sparse, resulting in poor 3D object detection in the unmanned environment perception system.Multi-frame point cloud registration can achieve sparse point cloud densification, however, pedestrians and moving vehicles in a dynamic environment will reduce the positioning accuracy of the LiDAR,and will also cause a large offset of the point cloud on the moving object in the fusion frame.Aiming at the above problems, this paper proposed a multi-frame point cloud fusion algorithm in a dynamic environment, and used this algorithm to detect 3D objects in the real situation of park roads, which improved the 3D object detection accuracy of low-beam LiDAR.This paper used the dri-ving road condition data collected by 16-line and 40-line LiDAR to conduct experiments.The results show that the algorithm can enhance the density of sparse point clouds and improve the environmental perception ability of low-cost LiDAR.