Target Tracking Algorithm In Wireless Sensor Networks

Due to its rapid implementation ability, Ad-Hoc network architecture, and the advantage of fault-tolerant and high coverage, wireless sensor networks (WSNs) have achieved widely attention in the academic and industrial, and has been applied in lots of applications, such as animal habit monitoring, patient monitoring, battlefields surveillance, et al. The ability of tracking the target based on the WSNs is of vital importance for the above applications. However, due to the limited computational ability and battery energy, and the complex environment of the WSNs applications, how to realize the target tracking task efficiently and robustly becomes an essential and important problem.This dissertation focuses mainly on the study of the target tracking algorithms based on the WSNs under the Bayesian estimation theory. The target tracking algorithms that based on the distributed particle filter and the probability density propagation are proposed, and the tracking algorithms which are robust to the complex radio signal propagation environment are also presented. The main contributions of the dissertation are summarized as follows:Firstly, to make the traditional computational complex particle filter algorithm applicable to the resource limited WSNs, a novel collaborative tracking algorithm based on the distributed particle filter is proposed. The dynamic cluster based network architecture and schemes to select the cluster heads and the cluster member nodes are presented, and the collaborative scheme to realize parallel computation among cluster members are designed. Meanwhile, to overcome the particle degeneration problem, a particle distribution optimization scheme based on the current observation information is proposed.And then, to meet the high robustness requirement for the tracking algorithm in some applications, an algorithm based on the probability density propagation which can deal with the tracking task under non-linear, non-Gaussian, and strong noise circumstance is proposed. The Gaussian mixture model is adopted to represent the prior density distribution, posterior density distribution and likelihood distribution. The unscented transformation is used to deal with the non-linear prediction, and approximation method is used to achieve the posterior density distribution. Finally, the weighted centroid point of the posterior density distribution's different modes is calculated and set as the current position of the target.Finally, to overcome the effect of environment dynamic, a tracking scheme based on the dynamic radio map is proposed. The parameters of the environment and radio signal propagation model are dynamical monitored by the monitor nodes with the regression method, and the radio map is built based on the newest parameters received from the nearest monitor node. A particle filter algorithm is used together with the dynamic radio map to realize tracking. Meanwhile, for realizing robust target tracking in the circumstance where the radio signal propagation parameters are unknown, a novel tracking algorithm under the particle filter framework is proposed. Based on the monotonic relationship of the distance and the received signal strength, a lightweight scheme to realize particle weight calculation blindly is proposed.The above works enrich the algorithm theory of the target tracking based on the WSNs, and make a good foundation for applying the WSNs in numerous fields.