Average marginal density based distributed multichannel fusion for multi-target tracking

This paper proposes a novel distributed multi-target fusion mechanism based on average marginal density (AMD) via the generalized Covariance Intersection (G-CI) fusion algorithm. There exist several drawbacks in traditional multi-sensor tracking methods, e.g. track association is sensitive to the parameter; tracks fusion can not fuse multiple tracks jointly and traditional distributed fusion methods only apply to Gaussian distribution. To solve these problems, a robust distributed fusion method for multi-target is proposed in this paper. Firstly, we approximate the local multi-target posterior as a product distribution with its AMD which is proved to be the minimized Kullback-Leibler divergence of local multi-target posterior. Secondly, considering the unknown correlation between different sensor nodes, the G-CI rule is employed to perform distributed fusion. Since the track association process is embedded in G-CI fusion, the distributed fusion performs the tracks association and tracks fusion in company. Finally, we derived the closed-form solution of G-CI fusion with AMDs. The proposed fusion algorithm is implemented using Gaussian mixture and its performance is highlighted by numerical results.