基于粒子滤波的多自由度运动目标跟踪

摘要

为了在复杂背景下跟踪视频序列中的多自由度运动目标,基于粒子滤波理论提出了一种多自由度运动目标的稳健跟踪算法.首先,采用均值漂移算法目标模型与候选模型的相似度作为观测值的构造基础;然后,在核函数下颜色直方图的基础上,对目标的中心位置和表征目标形状的协方差矩阵进行更新,从而自适应地调整核函数带宽的大小,修正跟踪窗口的尺寸,实现对多自由度运动目标的跟踪,在粒子滤波中,取粒子数N为100,目标参考区域中心在x,y轴上的坐标分量随机游动的方差为5,参考区域在x,y轴上的尺度及角度随机游动的方差为0.1;在不同场景和不同目标的跟踪实验中,提出的算法能够稳健、可靠地跟踪多自由度运动目标,对目标尺度和角度变化具有良好的适应性.%In order to robustly track the multi-degree-of-freedom moving objects in video sequences at a complex background, a tracking algorithm for multi-degree-of-freedom moving objects was proposed based on the particle filter principle. Firstly, the similarity of a target model and a candidate model was taken as the structural basis of observation by using mean shift algorithm. Then, based on the kernel-color histogram, the center position of the object and the covariance matrix that described the shape of the object were updated to adjust kernel-bandwidth and modify the size of tracking window,then to implement the tracking for multi-degree-of-freedom moving objects. In particle filter, the number of particles is to be 100, the variance of coordinate components is 5 in the covariance matrix,and the variance of scale and angle components is 0.1. Tracking experiments for various objects in different scenarios show that the proposed algorithm can track multi-degree-of-freedom moving objects steadily, and can adapt to the change of scales and angles for objects.