摘要
针对粒子滤波方法在重采样阶段容易造成样本有效性和多样性的损失,导致了样本贫化问题,提出了一种改进的粒子滤波算法。算法将粒子群优化思想引入粒子滤波中,在粒子采样过程前先利用粒子群算法进行优化。粒子群算法将最新观测值融合到粒子进化公式中,大部分粒子经过粒子群优化后,朝着后验概率分布比较密集的区域运动,聚集在最优粒子附近,使粒子的权值被提高,避免了在重新采样过程中被舍弃,进而缓解了样本被贫化问题。目标跟踪系统中的位置估计由于物体运动具有突然性,很难准确估计。采用非线性目标跟踪模型和分时恒定值模型分别研究改进粒子滤波算法对误差均方值的影响。仿真结果表明改进算法与常规粒子滤波算法和扩展卡曼滤波算法相比,更加有效地降低变量的误差均方值,从而提高了滤波性能。
For the sample impoverishment problem of particle filter method, which is caused by the lost of the effectiveness and the diversity in the re - sampling process, a modified particle filter algorithm was proposed in the paper. The particle is optimized by the particle swarm optimization before the sampling process. Due to particle swarm optimization algorithm incorporated the latest observations into the particle evolution formulas, most of the particles after particle swarm optimization optimized moved to the dense area of posterior probability distribution and gathered in the vicinity of the optimal particle. As the weight of particle was bigger, the particle was not abandoned in the re -sampling process, so the sample impoverishment problem was alleviated. Because the motion was sudden, it was difficult to accurately estimate location of the object in target tracking system. The algorithm was tested using the nonlinear target tracking model and time - sharing constant value model. Compared with the conventional particle filter algorithm and the extended Kalman filtering algorithm, and the mean square error values of this algorithm is mini- mum. This suggests that the algorithm has better fihering performance.
出处
《计算机仿真》
CSCD
北大核心
2014年第8期392-396,共5页
Computer Simulation
基金
国家自然科学基金项目(11374001)
关键词
粒子群算法
粒子滤波
重要性采样
目标跟踪
Particle swarm optimization （PSO）
Particle filter
Importance sampling
Target tracking
