Study Of Target Tracking Algorithm In Wireless Sensor Networks

Wireless sensor networks are expected to find a wide variety of applications in many areas, including environmental monitoring, intelligent agriculture, target detection and tracking, robotics applications and so on. The technology of wireless sensor networks is expected to be embedded into the Internet, so as to facilitate information exchange between the external world and human beings, and broaden the way people seek information. Because of the merit of low power consumption, low cost, high capacity of self-organizing, scalability and imperceptibility, wireless sensor networks will play a more and more important role in the field of surveillance and tracking.This thesis gives an introduction of the knowledge of wireless sensor networks' components, architecture, routing protocols, application prospects and so on at the beginning. After that, some node localization algorithms frequently used in wireless sensor networks are described. Multidimentional scaling map algorithm is employed to achieve node localization in wireless sensor networks, and quantitative analysis depends on different simulation parameters is provided. Finally, the computational procedures and simulation analysis of extended Kalman filtering algorithm, unscented Kalman filtering algorithm and particle filtering algorithm is presented. Aiming at multisensor fusion based target tracking applications in wireless sensor networks, a mixed algorithm is proposed, called extended-mixed particle filter(EM-PF). The algorithm utilizes a mixed particle propagation scheme. In the process of multi-sensor measurement fusion using a particle filter in a wireless sensor network, a certain number of particles in the particle filter are propagated by using a Gaussian distribution obtained from an extended Kalman filter as the proposal distribution, while the rest of the particles are simply propagated by using the state transition prior distribution. Simulation results of fusion-tracking in wireless sensor networks show that, with a similar simulation speed, the mixed filter significantly increases the tracking accuracy and robustness as compared with the results obtained by the pure particle filter.