Tracking human walking in dynamic scenes

Extracting from a video sequence a representation for humans in motion has numerous applications. This task is difficult due to the complex nature of the human body which is non-rigid and capable of performing a wide variety of actions. We propose a model-based approach to tracking human walking in dynamic scenes. We model the human body as an articulated object connected by joints and rigid parts, and describe the human walking process as a periodic motion. The posture of the walker is determined by a recognition scheme that estimates the period and phase of walking. This result is then used to establish dynamic constraints for the human posture. These constraints along with kinematic constraints that govern the linkage of the articulated human body are then adopted to facilitate the tracking of the body parts of the human. The paper illustrates the results of testing our algorithm with real video.