基于激光点云坐标系的3D多目标跟踪算法研究

3D多目标跟踪是无人车领域中一个核心技术之一。3D多目标跟踪的相关工作多通过设计复杂的目标建模算法或数据关联算法,以寻求提高3D多目标跟踪系统的鲁棒性。为降低3D多目标跟踪系统的复杂性,将目标的三维中心点视为该目标进行跟踪,提出了基于激光点云坐标系的3D多目标跟踪算法。首先,使用3D目标检测器检测出每帧激光点云中的目标。然后,用卡尔曼滤波器预测上一帧目标的三维中心点在当前帧的位置状态,融合激光雷达坐标系下相邻两帧之间目标三维体积的交并比与目标三维坐标中心点坐标之间的欧式距离作为度量尺度,使用贪婪算法匹配最近邻的目标。在KITTI跟踪数据集上的实验结果表明,所提出的多目标跟踪算法表现优异,运行速度达到了63 f/s,且车辆类的sAMOTA达到了9432。

3D multi object tracking algorithm based on laser point cloud coordinate system

3D multi object tracking is one of the core technologies in the field of unmanned vehicle. In order to improve the robustness of 3D multi object tracking system,complex object modeling algorithm or data association algorithm are often designed. In order to reduce the complexity of 3D multi object tracking system,a 3D multi object tracking algorithm based on laser point cloud coordinate system is proposed,in which the 3D center point of the object is regarded as this object. Firstly,the 3D object detector is used to detect objects in each frame of laser point cloud. Then,the Kalman filter is used to predict the position of the 3D center point of the object in the current frame. The intersection and union ratio of the three dimensional volume of the object between two adjacent frames and the Euclidean distance between the coordinates of the three dimensional coordinate center point of the object are used as the measurement scale,and the greedy algorithm is used to match the nearest neighbor object. The experimental results on the KITTI tracking dataset show that the proposed multi object tracking algorithm has excellent performance,the running speed reaches 63 f/s,and the sAMOTA of cars reaches 94.32%.