改进的粒子滤波算法及其在车牌跟踪中的应用
Improved particle filter algorithm and its application in target tracking

针对粒子滤波算法精度、效率不高及样本贫化等问题,提出通过量子粒子群算法和自适应遗传算法改进的粒子滤波算法。在粒子滤波重采样之后,考虑采用量子粒子群算法的位置更新方程对粒子分布进行改善;再按适应度大小对样本排序,滤除适应度值低于平均水平的粒子,选取相应数量较优粒子替换被滤除粒子。为保证样本多样性和有效粒子数量,引入自适应遗传算法对粒子进行交叉、变异操作。选择非线性目标跟踪模型和分时恒定值模型对本文改进算法进行仿真,仿真结果表明本文算法精度、数值稳定性均高于同类算法;最后将本文算法运用于汽车视频跟踪实验中,实验结果表明本文算法对目标跟踪中物体快速运动、光线和背景剧烈变化的情况都有准确的跟踪效果。
To solve the problems of low accuracy, inefficiency and sample impoverishment of particle filter method, the improved method combining quantum particle swarm optimization and adaptive genetic algorithm was proposed. After re-sampling, the particle distribution was improved by the position renewal equation of the quantum particle swarm optimization. Then the samples were sorted according to their fitness, and the particles with fitness values less than average fitness were filtered. Then optimal samples were selected to replace the abandoned ones and crossover, mutate with adaptive genetic algorithm, so as to ensure the sample validity and diversity. The modified algorithm was simulated in nonlinear target tracking model and time-constant value model and proved to be high in algorithm accuracy and numerical value stability. This method was also applied to the car tracking experiment and proved to be very efficient and accurate especially under the condition that the target moved fast and the intensity and background changed dramatically