Bayesian approaches for target positioning/tracking in sensor networks

貝葉斯方法在傳感器網絡中的目標定位/跟蹤的應用

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

  • Hongqing LIU

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date2 Oct 2009

Abstract

Sensor networks have numerous remote monitoring and control applications, and target positioning/tracking is a fundamental and crucial issue in its operation and management. Bayesian approach is an advanced technique in applications of target positioning/tracking. In this thesis, a number of algorithms are devised for target po- sitioning/tracking in sensor networks under di®erent scenarios based on the Bayesian method. In some scenarios, non-line-of-sight (NLOS) distance measurements are present due to obstructions and sources of large positioning errors, and therefore they need to be considered. A Bayesian algorithm is developed without the line-of-sight (LOS) identi¯cation step based on factor graph. By recursively updating the position distri- bution with receiving messages, each sensor ¯nally gets target position estimate upon convergence. Simulation results show that the proposed algorithm is comparable to the Cram¶er-Rao lower bound (CRLB) of the sensor positions, in which perfect NLOS identi¯cation is assumed and only LOS distance measurements are used. In case of target tracking, particle ¯lter (PF) provides a numerical solution to the nonlinear and/or non-Gaussian Bayesian estimation problem. A target tracking al- gorithm is devised without measurement noise distribution information under NLOS propagation. The Lp-norm criterion is adopted to identify the LOS measurements. The algorithm is tested under three di®erent noise distributions, namely, Gaussian, Gaussian mixture, and ®-stable noises. Results show the e®ectiveness of the proposed algorithm. Due to the sensing range and energy consumption constraints on sensors, it is not practical to activate all sensors all the time. Therefore, sensor selection approaches are presented based on minimizing the posterior CRLB (PCRLB). The major contri- bution is to develop di®erent search strategies to select sensors. In some sensor networks, a centralized processing unit may be unavailable and thus centralized algorithms cannot be applied. A distributed PF (DPF) is developed for target tracking in this scenario. Support vector machine is used to compress particles to reduce the transmission energy. Two distributed averaging algorithms, namely, consensus ¯lter and gossip method, are adopted to fuse the local estimate from each sensor to obtain the global estimate.

    Research areas

  • Sensor networks, Bayesian statistical decision theory, Programming