Research On Multiple Extended Target Tracking And Trajectory Maintenance Algorithms Based On Random Finite Set

With the continuous progress of the modern sensor technology,extended target tracking(ETT)is becoming one of the most active research fields in information fusion,and has been widely used in a series of military and civil fields,such as Aeronautics,Astronautics,Robot navigation,Vehicle tracking,etc.The typical single-point target tracking only uses the motion information to track targets,which leads to the low rate of information utilization of high precision sensor systems.Therefore,the extension states,such as shape and size,should be considered to improve the information utilization of modern high precision sensor systems and achieve a better tracking performance.Based on the random finite set(RFS)theory,this dissertation carried out a systematic and in-depth research in ETT,and focuses on solving the ETT problems of shape estimation,measurement set partitioning,track association and tracking of non-ellipsoidal targets in complex situations.The main contributions of the dissertation are as follows:1.For the problem of shape estimation of extended targets,a novel shape estimation algorithm is proposed using B-spline curve fitting method.In the Extended target Gaussian mixture PHD(ET-GM-PHD)filter,the target extension information is assumed to be a given value,thus the ET-GM-PHD filter cannot provide the shape estimation of targets.To solve this problem,the proposed algorithm improves the ET-GM-PHD filter to estimate the target kinematical state,and meanwhile estimates target shape based on the measurements information using a B-spline curve,thus the proposed algorithm is called Shape-ET-GM-PHD filter.The simulation results show that,compared with the ET-GM-PHD filter,the proposed filter can estimate the shape of irregular extended targets without changing the tracking performance of the ET-GM-PHD filter.Thus the proposed method has a better prospect of application.2.For the problem of partitioning closely spaced targets in complex situations,a shape selection partitioning(SSP)algorithm is proposed.Compared with the ET-GM-PHD filter,the Gaussian inverse wishart PHD(GIW-PHD)filter can estimates target extended state using Random Mitrix(i.e.the covariance matrix of the target measurement set).However,the precision of partitioning algorithms used by the GIW-PHD filter will decline when targets are closely spaced,which leads to precision decrease of the GIW-PHD filter.To solve this problem,the proposed SSP algorithm uses the target shape information to divide the measurement set into cells,and then uses the likelihood function of the GIW-PHD filter to select the best division scheme,thus the partitioning performance of measurements generated by closely spaced targets will be improved.The simulation results show that,when two targets are closely spaced,compared with the standard GIW-PHD filter,the GIW-PHD filter using the SSP algorithm will take more time costs but achieve better performance.3.For the problem of track association of multiple extended targets in complex situations,a novel track association algorithm using target shape information is proposed.In most of the tracking systems in the real world,providing the trajectory information to users is a basic function.However,the standard extended target PHD filtering framework cannot provide target trajectories,thus track association of extended targets is a significant research topic.The track association approaches for typical single-point targets PHD filter only use the target kinematical information to associate data,thus the association errors will occur when targets move closely in successive scans.To solve this problem,the proposed algorithm will save the target shape information by creating a shape table(ST),and then the target tracks will be matched by the shape information in ST.The simulation results show that,compared with the typical track association algorithm,the accuracy of the proposed algorithm is significantly higher than that of typical association algorithms when the target shape difference is obvious.4.For tracking of irregular shape extended targets,a non-ellipsoidal model GIW-PHD(NEM-GIW-PHD)filter is proposed.The standard GIW-PHD filter is based on an assumption that the target shape is ellipsoidal,thus its precision will decrease when target shape is non-ellipsoidal.To solve this problem,the proposed NEM-GIW-PHD filter is based on a non-ellipsoidal measurement model to improve the likelihood function of the standard GIW-PHD filter,and use the B-spline curve fitting method to estimate the target shape.Finally,the measurement set of non-ellipsoidal targets is partitioned by a modified SSP algorithm.The simulation results show that the performance of the proposed NEM-GIW-PHD filter is better than that of the standard GIW-PHD filter,when target shape is non-ellipsoidal.