Joint cost minimization for multi-object tracking
Abhijeet Boragule, Moongu Jeon · 2017
Online multi-object tracking generally addressed in tracking-by-detection paradigm. Multi-object tracking is more complicated when a frame consists of abrupt motion and similar appearance of objects. The abrupt motion and similar appearance cause ambiguity to the cost function, which gradually reduces assignment performance of the tracker. In this paper, we exploit the object appearance and dynamics to minimize the joint probabilistic cost. We claim that the two-fold assignment approach resolves the incorrect assignment into correct assignment in data association. The detection-to-track and track-to-track technique exhibits the better performance with the well-known MOT challenge dataset.