City Tracker: Multiple Object Tracking in Urban Mixed Traffic Scenes

Multiple object tracking (MOT) in urban traffic is exciting since there is large variation of road user appearance. In order to overcome that variation while tracking the objects, a deep learning detection-based tracking framework is implemented to employ classification information and associate different objects. In this paper, a method called City Tracker is proposed for MOT task in urban mixed traffic using urban traffic dataset. City Tracker contains two components which are YoloV3 detection and DeepSORT tracking. This method compares favorably to a current feature-based tracker for urban traffic scenes and achieves up to 0.8989 in MOTP and 0.4265 in MOTA.