Human motion tracking using nonsynchronous multiple observations

This paper proposes a method of tracking a human object by using nonsynchronous multiple-viewpoint images. The proposed method tracks human forms efficiently by using a Kalman filter to integrate observed information which is obtained nonsynchronously from multiple viewpoints. The experimental system is composed of multiple observation nodes, which operate nonsynchronously to process the multiviewpoint images, a tracking node, which tracks the human figure, and the discovering node, which finds the human figure. The image features are matched to the tracking model in the observation node based on the prediction of the observed value which is sent from the tracking node. The image features matched to the model are sent to the tracking node, and the tracking model is updated. The image features which are not matched are sent to the discovering node to find a new human figure. With the proposed approach, it is possible to construct a large-scale tracking system while reducing the deterioration of processing efficiency and the redundancy among observations that occur in synchronous systems. The effectiveness of the proposed method is demonstrated by an experiment using real images.