j

Motion is a powerful cue to distinguish objects: in many tracking scenarios it is possible to discriminate the targets from each other by only looking at their motion patterns. Yet, most state of the art approaches to multi target tracking rely heavily on appearance to associate detections from frame to frame and often overlook motion cues. In this work, we propose a multi-object tracking framework based on motion dynamics which is capable of tracking alike objects or objects with similar appearance and recover missing data due to long occlusions