Automatic recognition of human walking in monocular image sequences

In numerous content-based video applications, it is important to extract fr om a video sequence a representation for humans in motion. This task is dif ficult, because humans are not rigid objects and they are capable of perfor ming a wide variety of actions. However, often, human movements can be cate gorized into repetitive and rhythmic patterns of motion. Identifying the mo tion pattern of a human significantly alleviates the task of construction o f its representation. We propose here a model-based recognition of the gene ric posture of human walking in dynamic scenes. We model the human body as an articulated object connected by joints and rigid parts, and model the hu man walking as a periodic motion. The recognition task is to fit the model walker sequence to the walker in the live video (data walker sequence). We achieve this by determining the period of the data walker sequence and find ing its phase with respect to the model walker sequence. We present promisi ng results of how our system performs with a live video sequence.