Multiple Object Tracking in Urban Traffic Scenes with a Multiclass Object Detector

Multiple object tracking (MOT) in urban traffic aims to produce the\ntrajectories of the different road users that move across the field of view\nwith different directions and speeds and that can have varying appearances and\nsizes. Occlusions and interactions among the different objects are expected and\ncommon due to the nature of urban road traffic. In this work, a tracking\nframework employing classification label information from a deep learning\ndetection approach is used for associating the different objects, in addition\nto object position and appearances. We want to investigate the performance of a\nmodern multiclass object detector for the MOT task in traffic scenes. Results\nshow that the object labels improve tracking performance, but that the output\nof object detectors are not always reliable.\n