This paper studies the target tracking problem over wireless sensor networks (WSNs). While most existing works on this problem focus on the single-target case and the few existing two-target tracking methods are based on linear observation models, we pay attention to the two-target tracking over WSNs, considering nonlinear targets’ dynamics and observation models. We divide all the sensors in the WSN into two groups corresponding to the two targets such that sensors in each group only observe one target but collect the information of the other one by communicating with the sensors in the other group. Then, the consensus-based distributed two-target tracking (CDTT) algorithms are proposed, applying the covariance intersection fusion rule to the information form of the extended Kalman filter. With this fusion rule, the consistency of the algorithms is guaranteed and the estimation accuracy is improved. Through analyzing the boundedness of the information matrices and the Lyapunov candidate defined on the estimation errors, we prove that the two-target tracking task can be completed by running the CDTT algorithms under certain conditions. Moreover, we extend the CDTT algorithms to the multi-target case, obtaining the consensus-based distributed multi-target tracking algorithms and showing their performance analysis. Simulation and experimental results are given to illustrate the performance of these tracking algorithms.

本文研究了无线传感器网络（WSN）上的目标跟踪问题。虽然针对该问题的大多数现有工作都集中在单目标情况下，并且现有的少数双目标跟踪方法是基于线性观测模型的，但我们关注 WSN 上的双目标跟踪，考虑非线性目标的动力学和观测模型. 我们将 WSN 中的所有传感器分成与两个目标相对应的两组，使得每组中的传感器只观察一个目标，但通过与另一组中的传感器通信来收集另一个目标的信息。然后，提出了基于共识的分布式双目标跟踪（CDTT）算法，将协方差交叉融合规则应用于扩展卡尔曼滤波器的信息形式. 该融合规则保证了算法的一致性，提高了估计精度。通过分析信息矩阵的有界性和估计误差上定义的Lyapunov候选项，我们证明了在一定条件下运行CDTT算法可以完成双目标跟踪任务。此外，我们将CDTT算法扩展到多目标案例，获得了基于共识的分布式多目标跟踪算法并展示了它们的性能分析。给出了仿真和实验结果来说明这些跟踪算法的性能。