Research On Object Tracking Algorithms Based On The Distributed Sensor Networks

With the rapid development of the WSNs,various distributed algorithms have gradually raised the focus.The object tracking has always occupied an important role in some fields,such as military,navigation and so on.Therefore,the research about the distributed object tracking problem is vital.Distributed object tracking algorithms in the nonlinear system have received a lot of attention in recent years due to the value in the practical application.With the development of sensors,distributed extended object tracking algorithms also have huge potential value in some emerging fields,such as maritime supervision and unmanned driving.The distributed object tracking algorithm in the nonlinear system,the distributed extended object tracking algorithm,as well as the fusion methods and adaptive combination coefficients,which could affect the tracking performance of algorithms,are discussed in this thesis.The specific content is as follows:(1)Aiming at the problem of distributed object tracking in the nonlinear systems,this thesis based on the Bayesian framework approximates the complex exact local posterior probability density function with a simple Gaussian probability density function from the Kullback-Leibler(KL)divergence point of view.Via the diffusion strategy,the local posterior approximate probability density function could be obtained by the localized adaptation phase followed by the combined phase.For this reason,two localized adaptation algorithms based on two kinds of neighborhood KL divergence,and three combi-nation methods are discussed in this thesis.In the experiment,two localized adaptation algorithms are respectively combined with three combination methods.Accordingly six distributed object tracking algorithms in the nonlinear syetem could be obtained based on the KL divergence minimization.And these algorithms could be proved to effectively track the object via the simulations.(2)Aiming at the problem of the extended object tracking in distributed networks,since the ellipse shape is common in life and can be simply determined by three parameters,i.e.,direction,semimajor and semi-minor axes,the extended object is modeled as an ellipse shape in this dissertation.And the measurement obtained at each node is modeled by the multiplicative noise.With this shape model and the measurement mdel,a distributed extended object tracking algorithm based on the diffusion strategy is discussed in this dissertation.Herein,the kinematic state and the extension of the extended object could be respectively estimated via two distributed Kalman filters.The simulation results prove that the discussed distributed extended object tracking algorithm could effectively track the extended object.(3)It is necessary to consider both the combination of the kinematic state and the combination of the extension for the extended object tracking,where the diversity of the ellipse shape parameters may introduce some new problems.The application of the three combined methods for the distributed point object tracking are further explored to the distributed extended object tracking algorithm.Aiming at the combination problem of the distributed extended object tracking,an extended matrix combination method and the utilization of the adaptive combiner in the extended matrix are studied in this dissertation.The simulation results demonstrate that the improved tracking performance could be obtained by the extended matrix combination method and the adaptive combiner algorithm.