Study On Particle Filter Algorithm And Its Implementation For Radar Target Tracking

Tracking radars for non-cooperative targets have been widely used in both military and civilian applications. Tracking algorithm is the core technique for tracking radar, because the robustness of the algorithm directly affects the performance of tracking radar. The traditional tracking algorithms based on Kalman filters are usually not satisfied because of the nonlinear movement usually presented by non-cooperative targets in real situations. In recent years, particle filter based on sequential Monte Carlo method and recursive Bayes estimation has been highly focused due to its excellent performance in nonlinear and non-Gaussian noise situations, however, the huge amount of calculation and complex algorithm structure prevent it from being applied to real systems. The objective of this thesis is to conduct research on the algorithm and its hardware realization of particle filter for real radar system in view of tracking non-cooperative targets. The major works are summarized as follows:We firstly study and improve the action sequence and architecture of the sampling, weighting calculation and resampling steps based on the essential of particle filter theory, to make the particle filter much more fitted for parallel and pipeline operations, and then apply them to bearing only tracking system and construct a close-loop tracking system on the platform of FPGA Xilinx Virtex-5. Experiment results show that the improved architecture saves 50% of hardware resources, much better real-time performance, much more efficient and much simpler. Aiming at the degenerate problem of particles, resampling can be used to deal with it. The basic idea of resampling is to copy the particles with large weight and abandon the particles with small weight. We propose a new simplified resampling method based on comparison of several resampling methods, which is more suitable for hardware implementation using adaptive threshold of the weights. The new method can on one hand guarantee the number of particles unchanged before and after resampling, on another hand keep the accurate performance of the filtering and at the same time reduce the hardware complexity remarkably.In consideration of real applications, the thesis analyze the difference between predicting algorithm and tracking algorithm for particle filter on aspects of principle, action steps, computational complexity, hardware resource occupation, and so on. Aiming at the issue of echo pulse prediction for real radar system, take uniform linear motion for example, we effectively predict the valid position of the echo pulse by utilizing the prediction capability of particle filters. We propose and realize two close-loop tracking schemes based on the platform of FPGA provided by Xilinx so as to validate our design.