Research On Nonlinear Filtering Based Target Tracking Algorithm For Sensor Networks

Wireless sensor network is one of the most important technologies in the21st century. It’s also an emerging platform that supports remotely locate and track targets. Compared with traditional platforms such as global positioning systems, wireless sensor networks have the advantages as low cost, high density, wide spread, and high accuracy. It is more competent for real-time monitoring such as those under complex terrain and unattended environment. Specially, the advantages of wireless sensor network are particularly significant in target tracking applications. However, for target tracking in sensor networks, excessive ineffective or redundant information will increase the communication cost and make the tracking algorithm more complicated. These issues are fatal flaws for energy efficiency and tracking accuracy. Furthermore, information collection and processing also plays a crucial role in system resource consumption. Factors such as energy efficiency should be taken into account. Therefore, the research of precision and energy-efficient sensor network target tracking system has great significance. In the actual application environment, since the node measurements and target states tend to follow a nonlinear relationship, this paper proposed three kinds of nonlinear filter based target tracking algorithm for wireless sensor networks:(1) Target tracking algorithm based on dynamic cluster routing optimization and distributed particle filter utilizes a dynamic approach to divide the nodes which deployed randomly in the monitored region into a number of clusters, then optimizes not only the communication routes between member nodes and cluster head in each cluster, but also the routes between cluster heads and the base station. This scheme can effectively reduce the total consumption of sensor networks, achieve the goal of tracking and guarantee the tracking accuracy simultaneously.(2) Particle Swarm Optimization and Metropolis-Hasting sampling Particle Filter based target tracking algorithm adopts dynamic topology of the network structure and distributed algorithm. Particle Swarm Optimization and the Metropolis-Hasting sampling are introduced into the resampling period to solve the problem of sample impoverishment. In order to achieve the goal of high-precision tracking performance, the history informations are shared between the particles to reduce the correlation between the history states of a single particle, so that the particles can rapidly converge to an optimal distribution.(3) In the Dynamic spanning tree and improved Unscented Kalman Filter based target tracking algorithm, a tree network structure that can dynamically span and adjust itself with the motion of target is implemented as a platform for executing target tracking algorithm. In order to accelerate the computing speed of the target tracking algorithm, particle swarm optimization is utilized to improve the approximation process of state vector probability density function in Unscented Kalman Filter. The optimized distribution of δ sampling points is closer to the real target state, and thus the filter accuracy is improved.The above are wireless sensor network target tracking algorithms based on nonlinear filters. The target tracking problem is solved by combining the network structure and nonlinear filtering algorithm. In this peper, we make specific simulation examples to examine the performance of the above three tracking algotithms. The simulations corroborate that these algorithms can effectively reduce network energy consumption and guarantee the tracking accuracy.