An Arithmetic Geometric Mixed Average GM-PHD Algorithm for Decentralized Sensor Network With Limited Field of View

The nodes in a decentralized network have equal status and can communicate with their neighbors. Considering the target tracking problem for a strongly connected decentralized network with a limited field of view (FoV), each node can detect and track targets in its FoV, and all nodes need to achieve consistent tracking results after finite communications and fusions. In this work, we formulate the decentralized tracking and fusion problems as the corresponding probability hypothesis density (PHD) filtering and fusion problem. First, we reviewed the existing geometric and arithmetic average (AA) methods for the PHD fusion problem and proposed a mixed average fusion method that combines both geometrical and AAs. Then, we discussed the criteria for the three PHD fusion methods and demonstrated they can achieve consistent fusion results with sufficient communication. We use Gaussian mixtures (GMs) to approximate PHD in our implementation. As a result, the PHD fusion problem is transformed into a GMs’ fusion problem and can be effectively addressed through the proposed GM matching and fusing methods. Furthermore, we developed the decentralized GM-PHD algorithms. The simulation results showed that the proposed methods are effective and robust.