Data association in visual sensor networks based on high-order spatio-temporal model

One of the fundamental requirements for visual surveillance with visual sensor networks is the correct association of camera's observations with the tracks of objects under tracking. In this paper, we propose a high-order spatio-temporal model to deal with the problem of missing detection, and then formulate the data association problem with dynamic Bayesian networks. After presenting the exact inference algorithm for data association and showing its computational intractability, we derive two approximate inference algorithms based on different independency assumptions. To apply the algorithms when the object appearance model is unavailable, we incorporate the proposed inference algorithms into EM framework. Simulation and experimental results demonstrate the effectiveness of the proposed method.