Random Finite Set In Passive Radar Networks And Multi-target Tacking

Passive radars do not emit signals actively,but use the existing electromagnetic signals to detect and track targets in the surveillance area.Passive radars have many advantages over active radars in that they tend to be smaller and cheaper,have no electromagnetic pollution,and have the potential anti-stealth capability.To improve the detection and tacking performance of the radar system,the networking of multiple passive radars has been fully studied these years.However,there are a lot of difficulties in multi-target tracking in practice,such as the varying number of targets,miss-detections,sensor system noise,and clutters existing in the surveillance area.Besides,due to communication and real-time constraints,the passive radar network may need to select a subset of receivers to report high quality target-related measurements.This will further increase the difficulty of multi-target tracking.Traditional multi-target tracking methods usually use the data association technique and convert the multi-target tracking problem into multiple single-target tracking problems.The computational complexity can be huge when the number of targets and measurements is large.In recent years,multi-target tracking methods based on the random finite set(RFS)have attracted widespread international attention.These methods use the unified Bayes framework for filtering estimation of the overall multiple targets,and avoid the complex data association process thoroughly.Consequently,the RFS has opened a new direction for the study of the multi-target tracking problem.This dissertation focuses on the use of the RFS in the passive radar network and multi-target tracking algorithm.The main content and contributions are summarized as follows: 1.The receiver selection problem is studied using the cardinality-balanced multi-target multi-Bernoulli(CBMeMBer)filter.In the passive radar network,a method based on the multi-objective optimization is developed to solve the adaptive receiver selection problem in the process of tracking.In CBMeMBer,the posterior multi-target density is formed by the union of the multi-Bernoulli parameter sets for the legacy tracks and measurement-updated tracks.The legacy track and the measurement-updated track have different theoretical and physical meaning,and hence are studied separately in this dissertation.Specifically,two objectives are considered in the receiver selection process: 1)Maximizing the mean cardinality of the measurement-updated tracks.2)Minimizing the cardinality variance of the legacy tracks.These two objectives are conflicting and cannot be achieved simultaneously.Therefore,this problem is modelled as a multi-objective optimization problem and is solved using the Pareto method.The solving process is simple since both objective functions have analytical solutions.This enables a fast and efficient selection of receivers.2.The multi-sensor data fusion method is studied based on the CBMeMBer filter.Data associations are necessary in the traditional multi-sensor data fusion method.However,the clutters inevitably exist in the surveillance area of the passive radar network and the number of targets can be large.In this case,data associations require large amount of computing and it is difficult to meet the need of real time processing.Different from the traditional methods,three methods avoiding the data association are alternative when the CBMeMBer filter is used for tracking multi-target,namely random update,sequential update and parallel update.This dissertation introduces these methods in detail and compares their performances using simulation experiments with the bistatic range only measurement.3.The dissertation studies the use of RFS in the joint probabilistic data association(JPDA)filter.The JPDA is an effective method for multi-target tracking in the cluttered environment,however,it performs poorly in tracking of closely spaced targets due to the phenomenon of track coalescence.For JPDA,the posterior PDF can be described as a Gaussian mixture model(GMM)with each Gaussian component representing a data association hypothesis.To estimate the target state,the posterior PDF is approximated by a single Gaussian PDF at each time step.The accuracy of the state estimation can be improved by switching the posterior PDF,when the labels of the targets are irrelevant.The Kullback-Leibler divergence(KLD)is used as a measure of the dissimilarity between the posterior PDF and the single Gaussian PDF.The smaller the KLD is,the more accurate the Gaussian approximation is.Therefore,the posterior PDF is optimized to minimize the KLD,which will improve the tracking accuracy.Since the KLD is not tractable,an approximation of the KLD is used as the cost function in the optimization procedure.This cost function is a linear combination of multiple objective functions which are not conflicting.Hence,the minimization of the cost function can be obtained by minimizing all objective functions simultaneously.Theoretical analysis and simulation results show that the proposed method can effectively avoid the track coalescence problem and works efficiently.4.The dissertation studies the use of RFS in the joint integrated probabilistic data association(JPDA)filter.The JIPDA introduces the probability of target existence as the track quality measure and is an effective method for automatic multi-target tracking.At each time step,JIPDA approximates the posterior PDF by a Gaussian PDF to estimate the track states.However,the JIPDA also suffers from the track coalescence problem.When targets are closely spaced and their tracking gates overlap,the measurement originating from one target can be confused with another.Note that the covariance can influence not only the accuracy of the Gaussian approximation but also the size of the tracking gate,which are crucial for the state estimation.Therefore,the posterior PDF is optimized by controlling the error covariance to obtain better tracking performance.A cost function measuring the trace of the error covariance matrix is developed as the optimization criterion and an iterative strategy is used to minimize it.Theoretical analysis indicates that the size of the tracking gate can be reduced and the accuracy of the Gaussian approximation can be improved.Simulation experiments further verify the effectiveness of the proposed method.