基于FPGA的粒子滤波算法研究与实现

发布时间：2018-03-22 21:19

本文选题：粒子滤波　切入点：重采样　出处：《西安理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：粒子滤波是一种基于贝叶斯估计理论和蒙特卡罗方法的滤波算法,其基本思想是利用一组状态空间中的随机样本,来近似后验概率密度函数,最后根据蒙特卡罗方法对状态值进行估计。对于非线性,非高斯系统有着高度的适应性,且精度其逼近最优估计,成为当前最热门的滤波算法之一。本文首先对粒子滤波算法的研究现状和背景意义进行了介绍,然后从贝叶斯估计和蒙特卡罗方法入手,对序贯重要性采样算法做了详细的分析,并就粒子退化问题对重要性函数和重采样进行了介绍,在此基础上引出了基本的粒子滤波算法。由于粒子滤波算法需要大量的粒子参与运算,因此决定了其存在计算量大,复杂度高,实时性差的缺点,限制了粒子滤波在实时系统中的应用。因此本文基于重采样对粒子滤波算法进行改进和简化,介绍了常见的多项式重采样、系统重采样和残差重采样,并对其硬件实现的复杂度和运算速度进行了分析。为了进一步提高粒子滤波算法的实时性,本文基于部分重采样算法进行改进,提出了一种基于线性组合的重采样算法。改进的重采样算法电路结构相对较为简单,不需要进行权值归一化操作,并非简单地进行对大权值粒子进行复制,而是通过线性组合产生新的粒子代替小权值粒子,有效地避免了粒子多样性的丧失,这种算法在保证滤波精度的基础上,大幅度的降低了运行时间。最后本文以二维纯方位目标跟踪系统为应用场景,给出了改进简化后粒子滤波算法的明确步骤。同时对粒子滤波算法的硬件设计进行了研究,将其硬件结构主要划分为采样模块,权值计算模块,重采样模块和状态估计输出模块。分析了粒子滤波并行执行思路,通过将其分解为多个并行的PE单元,对粒子处理任务进行分解,从而提高算法的实时性。然后分别对各个模块进行Verilog硬件电路描述,并利用ModelSim进行功能仿真,验证了硬件电路的正确性。最后将各个模块进行连接,在FPGA开发板上完成设计验证。
[Abstract]:Particle filter is a filtering algorithm based on Bayesian estimation theory and Monte Carlo method. Its basic idea is to approximate the posterior probability density function by using a set of random samples in the state space. Finally, the state value is estimated according to Monte Carlo method. For nonlinear, non-#china_person0# system has high adaptability, and its accuracy approximates the optimal estimation. This paper first introduces the research status and background significance of particle filter algorithm, and then starts with Bayesian estimation and Monte Carlo method. The sequential importance sampling algorithm is analyzed in detail, and the importance function and resampling of particle degradation are introduced. On this basis, the basic particle filter algorithm is introduced. Because the particle filter algorithm needs a large number of particles to participate in the operation, it has the shortcomings of large computation, high complexity and poor real-time performance. Therefore, this paper improves and simplifies the particle filter algorithm based on resampling, and introduces the common polynomial resampling, system resampling and residuals resampling. In order to further improve the real-time performance of particle filter algorithm, this paper improves it based on partial resampling algorithm. A resampling algorithm based on linear combination is proposed. The circuit structure of the improved resampling algorithm is relatively simple. Instead of small weight particles, new particles are generated by linear combination, which can effectively avoid the loss of particle diversity. Finally, taking the two-dimensional azimuth only target tracking system as the application scene, the explicit steps of improving the simplified particle filter algorithm are given. At the same time, the hardware design of the particle filter algorithm is studied. The hardware structure is divided into sampling module, weight calculating module, resampling module and state estimation output module. The task of particle processing is decomposed to improve the real-time performance of the algorithm. Then the Verilog hardware circuit is described for each module, and the function is simulated by ModelSim to verify the correctness of the hardware circuit. Finally, each module is connected. Complete design verification on FPGA development board.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN713;TN791

【参考文献】