Decentralized Multi-Target Tracking for Netted Radar Systems with Non-Overlapping Field of View

In this paper, a robust and high-accuracy decentral-ized fusion strategy is proposed for multi-target tracking (MTT) in netted radar systems with non-overlapping field of view (FoV). Each radar in the network runs a local Probability Hypothetical Density (PHD) filter with the decentralized consensus protocol to reduce communication bandwidth and eliminate information inconsistency among nodes. In the above process, the most critical core is an effective fusion strategy. Our proposed method adopts the geometric covariance intersection (GCI) rule to improve fusion accuracy. However, the standard GCI fusion is not suitable for the netted radar systems with non-overlapping FoV because it only focuses on the targets within the intersection of radar FoVs. Consider that, we extend the weights in GCI fusion to be a set of state-dependent weights instead of scalars to perform GCI fusion in a more robust manner. Furthermore, the radar FoVs are always unknown and time-varying in practical scenarios. Towards addressing this case, we combine a clustering algorithm based on highest posterior density to maintain a good fusion performance. The Gaussian mixture implementation of the proposed method is provided. Numerical simulations are designed to verify the effectiveness of the proposed method.