Consider a video sequence captured by a single camera observing a complex dynamic scene containing an unknown mixture of multiple moving and possibly deforming objects. In this paper we propose an uns
