Optimizing traffic management in Danang: a comparative study of multi-object tracking techniques for real-time vehicle flow monitoring

Abstract

This study evaluates the effectiveness of various detection-based object-tracking algorithms to optimize accuracy and efficiency in traffic flow monitoring. Due to its high accuracy in detecting objects, YOLOv8 was chosen as the vehicle detector for this research, where precise and rapid vehicle detection was critical. Regarding object tracking, our focus centered on the evaluation of five prominent Multiple Object Tracking (MOT) algorithms, including BoTSORT, ByteTrack, DeepOCSORT, OCSORT, and StrongSORT. We introduce a comprehensive traffic urban dataset collected from intricate street networks in Danang City. Our experimental results show that the system has practical applicability in urban traffic monitoring. Notably, the best model achieves a detection accuracy of 0.721 on [email protected], and the High Overlap Tracking Accuracy (HOTA) surpasses 72% for tracking performance across diverse traffic scenarios. This shows the applicability of MOT algorithms and provides a detailed view of traffic flow monitoring, especially in Danang City, Vietnam.

Keywords

• traffic flow monitoring
• object detection
• yolov8
• multiple object tracking (mot)