Multiple Hypothesis Tracking with Kinematics and Appearance Models on Traffic Flow for Wide Area Traffic Surveillance

Abstract

This paper presents a study of multiple hypothesis tracking (MHT) of vehicles recorded in wide area motion imagery (WAMI) that has persistent coverage. To take advantage of visual information contained in such aerial imagery, the authors propose a novel MHT-KAM method that combines multiple hypothesis tracking (MHT) with a kinematics and appearance model (KAM). Experiments were designed and implemented to test MHT-KAM on synthetic data sets with various frame rates, traffic configurations, and detection error rates. The experimental results indicate that this method can achieve promising performance for tracking individual vehicles, even in saturated traffic flow. The experimental findings indicate that the combination of applying high appearance weights in MHT-KAM and using large Mahalanobis distance-based gating solves the longstanding “closely-spaced targets” problem. The results also reveal satisfactory performance on existing aerial imagery data sets with limited quality and frame rates. This novel MHT-KAM method combined with previous computer vision-based approach has the potential to achieve a reliable and robust traffic surveillance system for extracting accurate microscopic data from persistent WAMI for diverse applications.