Assessment of Vision-Based Vehicle Tracking for Traffic Monitoring Applications

Abstract

Multiple-target tracking (MTT) is one of the main components of traffic monitoring systems directly responsible for measuring traffic information. However, assessment and evaluation of these vehicle tracking systems greatly varies and are often incomparable due to different metrics and datasets. This paper focuses on comparing and assessing the viability of online multiple vehicle tracking systems for use in real-time traffic monitoring. Most online vehicle tracking framework uses background subtraction for real-time vehicle detection and various blob-based appearance models for real-time multiple vehicle tracking. The results show that commonly used metrics such as multiple object tracking accuracy (MOTA) and multiple object tracking precision (MOTP) are not necessarily reflective of traffic monitoring performance, particularly in terms of vehicle count accuracy. Furthermore, the track identity switching (IDS) metric is identified to significantly affect the vehicle count accuracy, particularly in terms of count precision, having a correlation coefficient of $r(100)=-0.709$ with P-value $p .