Smart Sensor Surveillance Networks, Target Tracking Algorithm

Particle filter method achieves recursive Bayesian filter via Monte Carlo simulation.It is simple,flexible and easy to be implemented,and has parallel structure.This method is suitable for nonlinear target tracking which is hard to solve by traditional algorithm.Its precision can approach optimal estimation.Unattended sensor technology used in Smart Sensor Network for remote battlefield surveillance applications requires state-of-the-art algorithms to address the unprecedented challenges faced in target tracking.These sensors are not able to distinguish individual targets,decide how many distinct targets in the range.This dissertation studies on particle filter and its implementation in target tracking in Smart Sensor Network.The studies focus on the distributed tracking algorithm,bearing-only target tracking,and the multi-target tracking in Smart Sensor Network.The details and results of studies involved in this paper are follows:(1) The maneuvering models and the measurement models for target tracking are discussed deeply and systemic.First,the maneuvering models suitable for ground target tracking are presented.Several tracking algorithms,including Kalman Filter and grid-based filter,are summarized,and their principle and applications are also discussed.Based on the analysis of the Bayesian Importance Sampling,sequential important sampling is presented as the basic of particle filter.The degeneracy problem and its reason are discussed further,and some improved algorithms to solve this problem are introduced as the evolution of particle filter.The simulation shows particle filter can solve the nonlinear problem in maneuvering target tracking, compared with extended Kalman filter which is widely used.(2) Considering the limitation of power and communication of the node in Smart Sensor Network,a novel distributed target tracking algorithm,distributed unscented particle filter,is proposed.Unscented transformation is used to generate the proposed distribution in parallel with more accuracy in the algorithm.It uses a deterministic sampling approach to get the estimation of the nonlinear function(accurate to the second order of the Taylor series expansion) Due to the unscented transformation, distributed unscented particle filter can enhance the efficiency of the particles,thus the amount of the particle can be small.Also unscented transformation can avoid the degeneracy problem in particle filter.Simulation shows distributed unscented particle filter can get more accurate tracking result with fewer particles and less communication,which in turn reduces the power consumption.(3) Considering the highly nonlinear tracking problem,bearing-only target tracking,a neural network aided unscented particle filter is proposed.First,the reason of missing target in unscented particle filter is analyzed in bearing-only target tracking, which is the SNR is changing quickly when the direction of angle is detected by the node.Based on the result above,neural network is introduced into the unscented particle filter.Neural network is able to learn and remember the tracking result.This helps unscented particle filter to get a better proposed distribution to match the SNR value.As demonstrated in our simulation results,the neural network aided unscented particle filter can improve the tracking result and give lower RMSE than usual unscented particle filter with reasonable time cost.(4) Multiple targets tracking in Smart Sensor Network are deeply researched.A multi-target particle filter tracking algorithm based on prediction is proposed.The analysis on the multi-target tracking in Smart Sensor Network shows it is different with the traditional tracking.First,Multi-targets tracking only happens in local area while not in the whole area of the network.Secondly,the measurement received by the sensor node is mixed with the signals from multiple individual targets,and it's hard to be separated.Thus,the common multi-target tracking algorithms can not used in this situation.The novel multi-target tracking method presented in this paper uses the predicting position,provided by particle filter,to separate the signal from the mixed measurement.Simulation experiments show it can solve the multi-target tracking problem,with only a few calculations added compared with single target tracking.