改进的多特征融合粒子滤波视频跟踪算法  

Improved Multi-feature Fusion Based Particle Filter Tracking Algorithm

作　　者：张钰婷[1] 王沛[1] 马燕[1] 张倩[1] 

机构地区：[1]上海师范大学信息与机电工程学院,上海200234

出　　处：《电视技术》2014年第19期47-50,共4页Video Engineering

摘　　要：视频运动目标的跟踪是一个典型的非线性、非高斯问题,粒子滤波是一个解决非线性、非高斯问题的主流方法,粒子滤波技术具有非线性等特性,在目标跟踪过程中得到了广泛的应用。传统粒子滤波跟踪算法的退化现象严重,经过几次迭代递推,权重方差随着时间推移而增大,为解决该问题引入均值漂移算法,调整初始粒子分布,使粒子集中于邻近的局部极大值区域内,以减少退化现象的发生。并且将颜色特征和边缘特征融合在粒子滤波跟踪算法中,在传统算法基础上提出改进,加入优化机制,使粒子的权值分布更加接近实际情况。实验结果表明了该算法的有效性。Video moving target tracking is a typical nonlinear non-gaussian question. Particle filter is a mainstream solution to solve nonlinear and non-Gaussian question. Particle filter technology has nonlinear characteristics, and it has become a hot research field of visual tracking. Traditional particle filter tracking algorithm severe degradation, after several iterative recursion, the weight variance increases over time. In order to solve the problem, mean-shift algorithm is introduced, it can adjust the initial particle distribution, make the particles concentrated in the neighboring region, and fuse the features of both color and edge in particle filter tracking framework. The traditional algorithm is improved, using more according with the model motion characteristics and the mean-shift clustering particles, etc. Experimental results prove the effectiveness of the improved algorithm.

关 键 词：粒子滤波 目标跟踪 均值漂移 特征融合 

分 类 号：TN911.73[电子电信—通信与信息系统]