基于Mean Shift的粒子滤波视频目标跟踪算法

Research on Video Target Tracking Based on Particle Filter and Mean-Shift

目标跟踪一直是计算机视觉领域的重要研究方向,由于拍摄环境的多变性、目标运动状态的不固定性、目标间的相互遮挡以及相似物的干扰等因素使视频目标跟踪变得更加复杂。本文以粒子滤波算法为框架研究目标跟踪算法,重点对目标特征提取、模型间相似度的度量方法、控制跟踪过程中粒子数量进行研究,以提高算法的鲁棒性、准确性和实时性。提出了一种基于Mean Shift的粒子滤波视频目标跟踪算法。该算法为了减少光照的干扰,采用HSV颜色空间的核函数加权颜色直方图来描述颜色特征,在粒子初始化和重采样后采用一次Mean Shift算法的迭代过程将一些作用弱的粒子不参与计算,以减少计算量。完成一次目标估计后更新目标模型,以适应目标在运动过程的变化。对比实验结果表明,本算法具有较好的实时性。

Target tracking has been already a research direction in the field of pattern recognition and computer vision. Especially, video target tracking has become a focus of study. Target tracking plays an important role in the intelligent monitoring, traffic monitoring, traffic statistics, and so on, due to the variability of shooting environment, the fluid of target motion state, the shading of targets and the interference of likeness, etc. These factors make video target tracking more complicated. Video target tracking can be defined as select single or multiple target from the video sequences filming with cameras, and can give target location accurately and timely, and then get target motion trajectory and its motor habit. Mainly in the framework of particle filter algorithm, this paper analyzes the factors which affect the effect of particle filter tracking, and focuses on the study of the target feature extraction, the similarity measurement method between models, controlling the number of particles, in order to improve the robustness, accuracy and instantaneity of the algorithm. Particle filter video target tracking algorithm based on Mean Shift is proposed. The algorithm uses an iterative process of Mean Shift algorithm after the particle initialization and resampling to reduce the amount of calculation. The algorithm updates the target model after estimating location of the target, in order to adapt to changes of the target. The experimental results show that the effect of the particle filter tracking based on Mean Shift is better, and it needs shorter time.