A Real-Time Multiple Object Tracking Network Under Complex Traffic Environment

Abstract

Multiple object tracking is extremely important in autonomous driving. Most multiple object tracking methods include two steps: object detection and data association. The first step is to detect each frame object in the video. The second step is to establish the relationship between video frames and to obtain the object trajectory. In this study, the M-YOLOV5s is proposed to train the improved BDD100K datasets based on the YOLOV5s. Then, the Hungarian algorithm is used to match the trajectory predicted by Kalman filtering with the detection object. The corresponding objects between the video frames are found and the multiple objects are tracked in real-time by constructing the similarity matrix. The experiment results on the MOT16 benchmark show that the whole algorithm has strong robustness and can achieve multiple object tracking under complex traffic environments.