Study Of Multi-target Tracking Algorithm Based On RFS And Its Application In Passive Radar

Tracking multiple moving targets accurately is a key technology in many fields.Because of the time varying of target number and measurement uncertainty,the multi-target tracking becomes a difficult problem in the field of target tracking.The traditional methods for solving the problem of multi-target tracking are to reduce this problem by using hard-association or soft-association technique into a set of single target tracking problems.The traditional methods do not establish a unified theoretical framework for multi-target,and are often unable to cope with complex scenes when the number of targets is unknown and time-varying.The multi-target tracking methods based on the random finite set provide an elegant and novel way to multi-target tracking.Based on finite set statistics,the multi-target tracking problem can be formulated in a Bayesian framework,which is an extension of the single target Bayesian filter and avoids the data association.One of the important functions of the passive radar is to achieve effective tracking of multiple moving targets simultaneously.Positioning with highly coarse measurements from multiple transmitter-receiver pairs in the passive radar will greatly increase the difficulty of multi-target tracking.This dissertation focuses on the multi-target tracking algorithms based on random finite set and their applications in passive bistatic radar.The main research results of this dissertation are summarized as follows.1.The standard implementation of the probability hypothesis density(PHD)filter selects the state transition function as the proposal density function to generate the predicted particles,which results in a low efficient use of particles and a serious degradation.To solve the problem,a novel particle PHD(P-PHD)filter based on the cubature Kalman filter(CKF)is proposed.This method uses the current measurements by the CKF to generate the proposal density function and obtains the predicted particles states by sampling from the proposal density function,so that particle distribution is closer to the real multi-target posterior probability density function and the particle degradation is alleviated.Compared with the standard P-PHDfilter,the proposed method can greatly reduce the number of particles required in the filter.Compared with the P-PHD filter based on the unscented Kalman filter(UKF),the proposed method has a lower calculation,a better stability and a higher accuracy of tracking when the dimension of target state is high.Finally,the simulation results show the effectiveness of the proposed method.2.A novel particle cardinality-balanced multi-target multi-Bernoulli filter(P-CBMeMBer)based on the square-rooted cubature Kalman filter(SCKF)is introduced.This method select the SCKF to construct the proposed density functionso as to guide the predicted particles to move towards the high likelihood areas,resulting in the optimization distribution of particles and the improved tracking accuracy.The SCKF does not require the square-rooting operation on the error covariance matrix which can result in the deterioration of accuracy and the filter divergence or termination.The simulations show that the performance of the proposed method is equivalent to that of the P-CBMeMBer filter based on CKF.However,by avoiding the square-rooting operation on matrix,the proposed method has better stability.3.The improved P-CBMeMBer methods which use the current measurements to generate the proposal function of all the surviving particles can alleviate the particles degradation effectively,but also bring massive calculation.What's more,these methods exhibit a noticeable cardinality overestimation in the condition of high clutter rate.To solve this problem,an adaptive P-CBMeMBer filter algorithm is proposed.Firstly,this method obtains the predicted particle states by selecting the target state transition density function as the proposal density function.Then the copy time of each particle is reserved at the step of resampling to determine the quality of the particle.The states of the particles with more repetitions are maintained and other particles are refined by the corresponding measurements.In addition,the step of measurement-updatein theP-CBMeMBer filter is simplifiedin the proposed method.Simulations show that the novel method manages to alleviate the particles degradation problem without increasing the computational complexity seriously and can avoid the cardinality overestimation resulting from the abusing of mearsuements.4.A P-PHD filter method which can be applied in the system of passive bistatic radar is proposed.The P-PHD filter assumes that the information of the birth targets is known a priori.However,this assumption is no longer holds under the background of radar.When the birth target information is unknown,the P-PHD filter needs to consume a large amount of particles to detect whether there is a new target to appear at every moment.In order to optimize the distribution of the birth particles,the proposed method uses the current measurements and combines with the multi-station location principle of passive bistatic radar to look for the areas where the new targets are most likely to appear.What's more,the measurements corresponding to each transmitter are fully utilized at the step of update to improve the accuracy of tracking.Due to the poor accuracy of the azimuth of the target in passive bistatic radar,the azimuth angle is only used to optimize the distribution of the birth particlesin the proposed method.Simulation results show the effectiveness of the proposed method.