基于激光雷达的自动驾驶车辆多目标检测与跟踪  

Multi-object Detection and Tracking of Autonomous Vehicles Based on Lidar

作　　者：朱爱鑫 袁春元[1] 陶振兴 ZHU Aixin;YUAN Chunyuan;TAO Zhenxing(College of Mechanical Engineering,Jiangsu University of Science and Technology,Zhenjiang 212100)

机构地区：[1]江苏科技大学机械工程学院,镇江212100

出　　处：《计算机与数字工程》2025年第3期815-820,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:51575249)资助。

摘　　要：针对自动驾驶车辆复杂场景中的环境感知问题,提出了一种基于激光雷达的多目标检测与跟踪方法。首先将基于坡度和高度的地面分割方法相结合,剔除原始点云数据的地面点;其次对非地面点使用引入K维树的密度聚类算法进行多目标检测,并过滤明显不感兴趣对象,为每个目标添加L形拟合框;最后考虑多目标的非线性运动,选用联合概率数据关联和无迹卡尔曼滤波的组合,对检测到的多目标进行实时跟踪。在公开的自动驾驶KITTI数据集上进行验证,结果表明,论文所提的多目标检测与跟踪方法具有较高的可靠性。Aiming at the problem of environmental perception in the complex scene of autonomous vehicles,a multi-object detection and tracking method based on lidar is proposed.Firstly,the ground segmentation method based on slope and height is combined to eliminate the ground points of the original point cloud data.Secondly,multi-object detection is carried out using the density clustering algorithm introduced into the K-dimensional tree for non-ground points,and the objects of obvious uninterest are filtered,and a L-shaped fitting box is added to each target.Finally,considering the nonlinear motion of multi-objective,the combination of JPDAF and UKF is selected to track the detected multi-target in real time.The results show that the multi-object detection and tracking method proposed in this paper has high reliability after verification on the publicly available KITTI dataset for autonomous driving.

关 键 词：自动驾驶 激光雷达 聚类 联合概率数据关联 无迹卡尔曼滤波 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]