融合多传感器的行人识别和多目标追踪算法  

A pedestrian recognition and multi-target tracking algorithm integrating multiple sensors

作　　者：李京鑫 孙健[3] 罗家毅 吴海波 LI Jing-xin;SUN Jian;LUO Jia-yi;WU Hai-bo(Department of Mechanical Engineering,Maanshan Technical College,Maanshan 243031,China;College of Automobile and Traffic Engineering,Nanjing Forestry University,Nanjing 210037,China;School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]马鞍山职业技术学院机械工程系,安徽马鞍山243031 [2]南京林业大学汽车与交通工程学院,江苏南京210037 [3]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《激光与红外》2025年第1期138-144,共7页Laser & Infrared

基　　金：2020年安徽省高校自然科学研究重点项目(No.KJ2020A0929);2021年安徽省高校自然科学研究重点项目(No.KJ2021A1338);2023年安徽省高等学校自然科学研究项目(No.2023AH052789)资助。

摘　　要：车辆自动驾驶过程中不仅需要完成运动规划,实现行人识别,还需实现更高精度的多目标追踪。针对车辆自动驾驶行人识别响应速度慢、目标追踪精度差的问题,提出一种融合多传感器的行人识别和多目标追踪算法。采用Lattice算法进行路径规划,通过损失函数与碰撞检测得到最佳驾驶轨迹,将传感器检测到的障碍物位置转换到全局坐标系中,在全局栅格地图中以高斯分布绘制在地图中,通过阈值初定可视行人。并基于检测跟踪策略设计多目标追踪遮挡处理算法,实现自动驾驶车辆中被遮挡目标的运动估计。对多目标跟踪挑战数据集的定量、定性和消融研究,验证了该算法的有效性。实验结果表明,该算法能够准确估计遮挡过程中的目标运动,生成完整、高质量的运动轨迹。During the process of autonomous vehicle driving,it is not only necessary to complete motion planning and achieve pedestrian recognition,but also to implement higher precision multi-target tracking.A pedestrian recognition and multi-target tracking algorithm integrating multiple sensors is proposed to address the issues of slow response speed and poor target tracking accuracy in autonomous vehicle driving.Using the Lattice algorithm for path planning,the optimal driving trajectory is obtained through loss function and collision detection.The obstacle position detected by the sensor is converted to the global coordinate system,and Gaussian distribution is drawn on the global grid map.The visible pedestrian is initially determined through threshold.And a multi-objective tracking occlusion processing algorithm is designed based on the detection and tracking strategy to achieve motion estimation of occluded targets in autonomous vehicles.Quantitative,qualitative,and ablation studies on the multi target tracking challenge dataset validate the effectiveness of the algorithm.The experimental results show that the algorithm can accurately estimate the target motion during occlusion and generate complete,high-quality motion trajectories.

关 键 词：行人识别 多目标追踪 遮挡处理 算法 LATTICE 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] U461.91[自动化与计算机技术—计算机科学与技术]