基于中心点距离关联的三维多目标跟踪算法  

3D Multi-object Tracking Based on Center Distance Association

作　　者：董艳秋 万旺根[1,2] 王旭智[1,2] 孙学涛 

机构地区：[1]上海大学通信与信息工程学院,上海200072 [2]上海大学智慧城市研究院,上海200072

出　　处：《工业控制计算机》2023年第12期12-14,共3页Industrial Control Computer

摘　　要：三维多目标跟踪技术在自动驾驶中发挥着重要的作用,许多基于检测的三维多目标跟踪方法被提出。数据关联是多目标跟踪中的关键步骤,基于中心点的检测网络在三维多目标检测上的有效性得到了广泛验证。因此,充分利用检测框与跟踪框之间的中心点距离信息,在三维多目标跟踪网络中,提出了基于中心点距离交并比(DIoU)的检测预处理以及数据关联优化算法。在公开数据集nuScenes上的实验结果表明,该网络与其他方法相比,在自行车、汽车以及行人等类别的平均跟踪精度上有明显提升,在ID切换次数上有较好的抑制效果。3D multi-object tracking plays an important role in autonomous driving systems and numerous approaches under the"tracking-by-detection"paradigm have been proposed.Data association is a key step in multi-object tracking,and the effectiveness of center-based detection network in 3D multi-object detection has been widely verified.In this paper,the center-distance between detection bounding boxes and tracking bounding boxes is fully utilized.This paper proposes a 3D multi-object tracking algorithm with the DIoU-NMS based detection preprocessing and the DIoU based data association.The experimental results on nuScenes show that,compared with other methods,the algorithm has an improvement in the AMOTA of bicycles,cars and pedestrians,and has a restraining effect on the ID-switches.

关 键 词：三维多目标跟踪 数据关联 交并比 非极大值抑制 

分 类 号：U463.6[机械工程—车辆工程] TP391.41[交通运输工程—载运工具运用工程]