一种用于3D激光雷达点云处理的多目标跟踪算法  

A Multi⁃object Tracking Algorithm for 3D LiDAR Point Cloud Processing

作　　者：武宏伟 吕东升 贾琳 WU Hongwei;LYU Dongsheng;JIA Lin(Beijing Wanji Technology Co.,Ltd.,Beijing 100085,China;Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China)

机构地区：[1]北京万集科技股份有限公司,北京100085 [2]中国科学院合肥物质科学研究院,安徽合肥230031

出　　处：《信息与控制》2024年第4期508-519,共12页Information and Control

摘　　要：近年来,3D激光雷达作为一种高精度传感器,提升了城市交通的治理水平。然而城市交通场景复杂,目标众多且运动轨迹相互交叉,传统的多目标跟踪方法难以生成准确的交通目标运动轨迹。本文针对城市路口复杂场景交通目标实时跟踪问题,设计了一种基于多层级结构的多目标跟踪(multi-level object tracking, MLOT)算法框架,结合基于椭圆门限的多维特征数据关联以及针对预测结果的自适应初值滤波等方法。通过对路侧感知数据集进行实验,结果表明,本文的算法明显优于初始的跟踪算法,能够增强复杂交通场景中目标跟踪关联的准确性和鲁棒性,提高感知跟踪的准确率,具备一定的工程推广价值。As a high⁃precision sensor,3D LiDAR has improved urban traffic management in recent years.However,urban traffic scenes are complicated,with numerous objects and intersecting traj⁃ectories.This makes traditional multi⁃object tracking methods ineffective for accurately generating traffic object motion trajectories.In this study,the real⁃time tracking of traffic objects in complex urban intersection scenarios is addressed,and a multi⁃layer object tracking algorithm framework is proposed based on a multi⁃level structure.The framework combines multi⁃dimensional feature data association based on elliptical thresholds,adaptive initial value filtering for predicted results,and other methods.Experimental results using roadside perception datasets reveal that the proposed al⁃gorithm outperforms the initial tracking algorithm.The proposed algorithm enhances the accuracy and robustness of object tracking associations in complex traffic scenes and improves the accuracy of perception tracking.Thus,it has certain engineering application value.

关 键 词：多目标跟踪 轨迹置信度 数据关联 椭圆门限 目标运动模型 

分 类 号：TN958.98[电子电信—信号与信息处理] TP18[电子电信—信息与通信工程]