Research On Methods For Passive Multi-sensor Target Tracking Based On Random Finite Sets

The methods of multi-target tracking based on Random Finite Set(RFS) provide a new way for multi-target tracking. Multi-target tracking algorithms based on RFS deduced multiple targets tracking formula under unified frame of RFS theory, estimating the number and states of targets synchronously, avoiding data association, has wide application prospects. Thus, this dissertation focus on the problems of how improve the precision of PHD filter, multi-target tracking based on RFS with unknown clutter intensity and multi-sensor multi-target tracking based on multi-target multi-Bernoulli(Me MBer) filter. The major contributions are as follows:1. Studied the problem of how improve the precision of single sensor multi-target tracking based on PHD filter. First, applied the technologies of particle labeling to sequenctial Monte Carlo PHD(SMC-PHD) filter, and proposed an improved SMC-PHD filter based on particle label. The proposed method meliorated the precision of estimated target states of SMC-PHD filter according to particles’ label and likelihood between measurements and particle state. Second, from the point of view of particle sampling, a SMC-PHD filter based on Unscented Transformation is presented. In the filter, the problem of particle sampling was resolved with the method of unscentd transformation, and the precision of estimated target state was improved. At last, in order to improve the precision of multiple model PHD(MMPHD) filter, an improved MMPHD filter was proposed. In the proposed method, the information of measurements are adopted for estimated the probability of kinematic model, and distribute limited particles in the area where targets are likely been detected, consequently, improve the precision of mlutple model PHD filter.2. Studied the problem of single sensor multi-target tracking based on RFS with unknown clutter. Firstly, aiming the recent proposed PHD filter with unknown clutter, analyzing the influence of the parameters of clutter model to estimated clutter rate theoretically, and proposed a method of setting parameters of clutter model, improved the PHD filter with unknown clutter. At last, the multi-target Me MBer filter with unknown clutter intensity is studied. By incorporating the gating technique, an improved Me MBer filter with the intensity of clutter unknown is proposed. The propose algorithm divide the field of sensor view into many areas, and estimate clutter intensity of validation areas. Simulation results demonstrate that the proposed algorithms can estimate target number and state precisely and have great tracking capability.3. Studied the problems of multi-sensor multi-target tracking based on RFS. Firstly, a multi-sensor Iterated-corrector Me MBer filter is proposed by extending the Iterated-corrector PHD filter to multi-sensor multi-target tracking based on Me MBer filter, and analysed the performance of Iterated-corrector Me MBer filter. Simulation results show that when the performances of sensors are not uniform, the different order of measurement updating will result in different result of data fusion. Thus, a two sensors Me MBer filter is deduced theoretically and the SMC implemention is proposed. Simulation results show that the performance of the proposed method is irrelated to the order of updating, and outperform the Iterated-corrector Me MBer filter. Despite the precise multi-sensor Me MBer filter in theoritically is optimal approach to multiple sensor tracking, it cannot used in practice because of too complexity in calculation when sensor number exceed two. So, at last, according to the theroris of Bayesian, a multi-sensor multi-Bernoulli filter for engineering application is proposed, and the SMC implementation is given. Simulation results demonstrate that the proposed algorithm outperform other algorithm of multi-sensor target tracking based RFS.