Robust tracking of cyclic nonrigid motion

Cyclic motion underlies several human activities including exercising, running, and walking. Accurate tracking of such motion in video data helps in developing computer-aided applications such as gait analysis, person identification, patient rehabilitation, etc. This paper presents a set of novel techniques for tracking cyclic human motion based on decomposing complex cyclic motion into simpler motion components and introducing phase coupling between the components. The intensity of coupling is adaptively adjusted during tracking such that a strong coupling is triggered when self-occlusion occurs. In our experiments we use sequential Monte Carlo methods for tracking a walking human. We show that this adaptive phase coupling of component motions handles occlusion and self-occlusion with significantly improved accuracy while avoiding the limitations caused by a poorly trained dynamic model.