Implementation Of Improved Particle Filter Algorithm Based On FPGA

Particle filter, a filter method based on recursive Bayesian estimation and Monte Carlo, has a unique advantage in dealing with parameter estimation and state filtering problem in the non-Gaussian and the nonlinear system. It can be applied to target tracking, navigation, image processing etc, but these areas require a higher real-time filtering and estimation. Also, the disadvantages of complex algorithm architecture, enormous computations leads to slow speed, lower real-time and need more resources. Field Programmable Gate Array has a strong power of parallel computation and it’s flexible. This thesis aims to reduce the complexity of particle filter with better computation speed and optimize the hardware structure, in order to design a particle filter which is fast, credible and using less hardware resource.This thesis, at first, introduces the elementary and improved particle filter algorithm, and imports some measures to prevent the lack of particles, such as re-sampling strategy and particle smoothing techniques. It analysis the structure and working principle of the FPGA, then selects the Spartan-3a as target device to realize this filter algorithm. Secondly, it uses the FPGA SOC to implement the improved particle filter EPF with MCMC. Local parallel particle filter algorithm is put forward, it partitions the particles to different parallel model which carries through sampling, re-sampling, calculating weight and MCMC step independently, then the main model integrates predictive value form sub-model to estimate the status of system. And it optimizes these steps of re-sampling, calculating weight and MCMC algorithm on the software implementation, which greatly reduces the space and time complexity of algorithm. Finally, it integrates the hardware implementation and software of SOC to verify the effectiveness of the design. And it simulates the non-Gaussian and the nonlinear system in Spartan-3a to check the correctness of the advanced algorithm, in order to realize the improved filter in real-time systems at last.This thesis verifies the effectiveness of using FPGA SOC to realize the improved particle filtering algorithm. The simulation results show that this algorithm has a fast, reliable performance and demand fewer resource, can be widely applied in target tracking, navigation and guidance, image processing, parameter estimation, and other real-time systems.