作　　者： ; ;

机构地区： 广州工程技术职业学院

出　　处： 《计算机工程与设计》 2017年第7期1905-1910,共6页

摘　　要： 针对基于粒子滤波(particle filter,PF)的目标跟踪算法易产生样本贫化的问题,提出一种利用和声搜索算法(harmony search,HS)优化重采样粒子滤波的视频目标跟踪方案。采用高斯混合模型(Gaussian mixture model,GMM)对背景建模,在目标视频帧中执行粒子滤波,通过直方图匹配法为每个粒子分配权重;利用和声搜索算法生成新的粒子,通过放弃一部分粒子来提高样本的随机性;对粒子进行重要性重采样,根据粒子权重估计目标状态。在BoBoT和DTU数据集上的实验结果表明,所提算法对目标背景、缩放、遮挡和光照等变化具有较好的鲁棒性,相比其它较新的目标跟踪算法,该算法取得了更好的性能。 For the issues that the traditional target tracking algorithm based on particle filter（PF） has the sample impoverishment phenomenon,a visual object tracking scheme based on re-sampling particle filter and harmony search（HS） optimization（HS-PF） was proposed.The hybrid Gauss mixed model（GMM） was used to model the background,and histogram matching method was used to assign weight for each particle.The new particles were generated by the harmony search,a part of the particles were given up to improve the randomness of the sample.Importance re-sampling was made for the particle,the target state was estimated according to the weight of particles.Experimental results show that the proposed scheme has certain robustness to the background,scaling,occlusion and illumination.And it has better performance than serveral other objective tracking algorithms.

关 键 词： 目标跟踪 和声搜索算法 重采样粒子滤波 样本随机性 高斯混合模型

领　　域： []