Abstract:

In order to obtain the movement trajectory of the vehicle in the traffic video and provide the dynamic traffic information of the road, a long-term multi-object vehicle tracking algorithm based on the combination of Yolo3 object detection and KCF object prediction, associated historical trajectory prediction results and detection results is proposed. For the non-smooth phenomenon of vehicle trajectory acquired by machine vision, the original vehicle trajectory is smoothed and optimized by Savitzky-Golay filter. Before and after optimization of vehicle trajectory in the test scenes, the optimized trajectory improves the trajectory smoothness under the premise of retaining original vehicle motion characteristics, and the dynamic traffic information provided can better reflect real motion condition of vehicle.

Key words: multi-object tracking, object association, trajectory optimization, Savitzky-Golay filter

摘要：

为了获取交通视频中车辆的运动轨迹，提供道路动态交通信息，提出一种基于Yolo3目标检测和KCF目标预测相结合，关联历史轨迹预测结果和检测结果的长时间多目标车辆跟踪算法；对采用机器视觉获取的车辆轨迹非平滑现象，提出通过Savitzky-Golay滤波器对原始的车辆轨迹进行平滑优化。对比测试场景中车辆轨迹优化前后，优化后的轨迹在保留原有车辆运动特征的前提下，改善了轨迹平滑性，提供的动态交通信息更能反映车辆真实运动状况。

关键词: 多目标跟踪, 目标关联, 轨迹优化, Savitzky-Golay滤波器