基于概率密度估计改进粒子滤波的行人跟踪算法研究

摘要

利用粒子滤波实现行人跟踪是视频智能监控的主要方法之一,但粒子滤波的粒子退化问题尚未得到一个比较理想的解决方法.本文利用重采样后的粒子集,构造经验分布函数,用支持向量机估计状态的后验概率密度模型,再依据该模型采样,在保证粒子有效性的同时增加了粒子的多样性,从而克服粒子退化现象,并基于加权颜色直方图模型进行了行人跟踪仿真实验.实验结果表明,该方法能有效克服粒子退化现象,跟踪精度相对于标准粒子滤波算法得到了提高,且该方法无需对后验分布作高斯假设,为解决粒子滤波算法中的粒子退化问题提供了一种方法.%Realizing pedestrian tracking using particle filter is a main method of video intelligent monitoring.But the problem of particle degradation in particle filtering has not been got an ideal solution.In this paper,first,by using particle set after resampling to formation the empirical distribution function,the posterior probability density model was estimated based on support vector machines.Then sampling according to the model.The effectiveness and diversity of particles were guaranteed at the same time,the degradation of the particle was eliminated effectively.Finally pedestrian tracking simulation was done based on weighted color histogram model.Simulation results demonstrate that it can overcome the phenomenon of diversity loss and particle degradation effectively.Tracking accuracy was improved compared with the standard particle filter.And it is no need to make the assumption that posterior distribution is Gaussion,the algorithm improved provides a way to solve the phenomenon of particle degradation in particle filtering.