A Study Of Multi-target Tracking Based On Random Finite Set Using Radar

Multi-target tracking is of considerable interest due to its wide applications in military and civilian fields. As new targets may appear or disappear randomly in the surveillance region, tracking multi-target involves simultaneously estimating the time-varying number of targets and their states from a sequence of observation sets in the presence of data association uncertainty, detection uncertainty, and noise. The random finite set (RFS) approach is an elegant formulation of the multi-target tracking. Using RFSs to model the multi-target state and observation, the multi-target tracking problem can be formulated in a Bayesian filtering framework by propagating the posterior distribution of the multi-target state in time. Compared with the traditional association-based techniques, the explicit associations between measurements and targets are avoided in the RFS formulation. This dissertation mainly investigates multi-target tracking based on RFS using radar. The author’s major contributions are outlined as follows:1. A new extension of the sequential Monte Carlo multiple-model probability hypothesis density (SMC-MMPHD) filter is presented in order to track multiple high-speed maneuvering targets via passive coherent location (PCL) radar. The standard MMPHD filter assumes that the target birth intensity is known a priori, in PCL radar where the targets can appear anywhere in the surveillance volume and the velocity range is wide, this is clearly inefficient. The extension enables us to adaptively initialize the new targets at each scan using the localization results of the measurements which are far away from the current estimated multi-target states and to effectively track multiple high-speed maneuvering targets. The simulation results show that the proposed method is viable.2. The standard cardinalized probability hypothesis density (CPHD) filter is a promising algorithm for multi-target tracking. However, due to the assumption that the target birth intensity is known a priori, it cannot work well in the situations where targets can appear anywhere in the surveillance region. To solve this problem, a one-step initializing Gaussian mixture CPHD filter is proposed to adaptively initialize the newborn targets using the measurements far away from the current estimated multi-target states. Furthermore, Doppler information (DI) is used to initialize the velocities of newborn targets, and in the update step position and Doppler measurements are incorporated in a serial process. Simulations show that the proposed algorithm can effectively initialize newborn targets and improve the estimation accuracy of target number as well as the optimal subpattern assignment (OSPA) distance when compared to the existing algorithm.3. The original Gaussian mixture (GM) multiple model cardinalized probability hypothesis density (MMCPHD) filter is an effective approach for multiple maneuvering target tracking. However, its algorithmic complexity is considerably high as the number of motion models increases. In this paper, we propose a novel extension of the GM-CPHD filter. Based on the best-fitting Gaussian (BFG) approximation, the proposed method predicts the intensity function and cardinality in a single model form. The update step is thus implemented independently of target motion models. Compared to the GM-MMCPHD filter with Mtarget motion models, the computational complexity is reduced by the order ofO(M). Simulation results are presented to demonstrate the effectiveness of the proposed method.4. Considering the limitation of the well-known multiple model formulation of the RFS that the statistics characteristic of clutter is assumed to be known a priori, this paper proposes a new multiple maneuvering target tracking algorithm based on Gaussian Mixture Cardinalized Probability Hypothesis Density (GMCPHD) filter in the case of unknown clutter. The proposed method predicts the intensity function of actual target states by BFG approximation, which is independent of the target motion model. Then the closed-loop iteration procedure among the intensity function of actual target states, the mean number of clutter generators, and the hybrid cardinality distribution of actual targets and clutter generators is established. The simulation results show that the proposed method can effectively estimate the target number, target states and the mean number of clutters simultaneously.5. A Track-Before-Detect (TBD) algorithm is presented to jointly detect and track multiple fluctuating targets under passive multistatic radar system based on Multi-target Multi-Bernoulli (MeMBer) filter. Because the amplitude likelihood is uncertain due to the unknown mean Signal-to-Noise Ratio (SNR) of fluctuating targets, firstly a uniform priori distribution is assumed for the mean SNR corresponding to the envelope output, and a likelihood function is marginalized over the range of possible values. Based on this approximated likelihood function, the fusion centre uses all the amplitude measurements from each receiver transmitter pair to update the predicted Bernoulli components. Simulations show that the proposed algorithm can jointly detect and track multiple fluctuating targets effectively, furthermore, the performance is similar to the situation of the known mean SNR when the value of the mean SNR is higher than 9dB.