A novel algorithm is proposed to perform object tracking with multiple cameras in a distributed computing manner. Firstly, the maximum likelihood criterion is employed to model the multi-view tracking problem. Then, the likelihood function is factorized based on the conditional independent assumptions. Finally, the estimation of the target location in each camera view is obtained by using an efficient message passing mechanism. In our algorithm, image data from all cameras no longer need to be transmitted to and processed by a central unit. Meanwhile, by taking full advantage of image data from multiple cameras, the tracking algorithm is very robust to occlusion. Both qualitative and quantitative experiments have demonstrated the effectiveness of the proposed algorithm.