Researches On Algorithms Of Bayesian Multiple Sensor Target Tracking

With the continuous development of networking technology,multistatic and networking observations have become a significant trend in modern surveillance system for the reason that multiple sensor observations generally reduce estimation uncertainties.And how to fuse multiple sensor observations is of significance in such systems.Especially with the rapid improvement of sensor manufactory technology,more and more defense systems employ low-cost sensors for large scale networking observations.Classical approaches are limited in such situations where no fusion center is available.In addition,in practical scenario,the observation process is under imperfect detections,which result in false alarms that make classical Bayesain methods limited.The main efforts of this thesis tend to tackle the above problems.Specially,in this thesis,the problem of target tracking is viewed as the problem of sequential state estimation under perfect detections and imperfect detections.Under perfect detections,information theoretic consensus estimation of the posterior is analysized.And its connection with the multiple sensor data fusion based on covariance intersection is studied.Numerical results in a bearing-only tracking situation show the algorithm is practical.Then the problem of moving source state estimation under imperfect detections is studied in the random finite set framework.Based on updating likelihood function sequentially,the multiple sensor Bernoulli particle filter is implemented.Numerical results show the effectiveness of the algorithm in situations with imperfect detections.Finally,multiple target tracking based on labelled random finite set is investigated.Simulation results show the filter based on labelled random finite set could simultaneously estimate multiple target state and identities.