Adaptive object tracking using bayesian network and memory

This paper presents an adaptive object tracking method that integrates the cues from color likelihood and edge likelihood, and that adapts itself to abrupt appearance changing objects. We use a Bayesian network based multi-modal fusion method of color and edge information. To handle the cases of sudden appearance changes, occlusion, disappearance and reappearance of tracked objects, a memory model is also introduced. The proposed tracker has the following characteristics. First, multiple modalities are integrated in the Bayesian network to evaluate the posterior of each feature. Secondly, context factors are computed in order to select best object state. Finally, a memory-based appearance model is introduced to handle abrupt appearance changes. Our method is robust and versatile for a modest computational cost. Desirable tracking results are obtained.