改进的基于颜色的粒子滤波目标跟踪算法

提出了一种改进的粒子滤波算法,在遮挡情况下,能鲁棒地跟踪运动目标.该方法是把改进的颜色直方图结合到粒子滤波的观测模型中,并提出了一种判断目标遮挡的分块检测遮挡的方法.首先对传统的以核函数赋权值的方法进行改进,把目标中心附近的像素都赋予最大的权值,目标的边缘由于遮挡等原因采用指数分布赋权值;在遮挡检测时,提出了把跟踪窗分为左右两个子部分,分别计算相似性度量的方法,提高了遮挡检测的实时性和准确性;同时,该算法对旋转和尺寸的变化具有鲁棒性.实验结果表明,与基本的粒子滤波算法相比,提出的新算法能更好的处理目标跟踪中的遮挡问题.

Improved color-based particle filter algorithm for object tracking

A novel object tracking algorithm is proposed, which could robust track objects under occlusion. The proposed method integrated with improved color histogram in the measurement model of particle filter, and an occlusion detection method based sub-block detecting is proposed. Firstly, a novel weights method is given, which improved the kernel function. The method is that the pixels where near the centre of object were given biggest weights, and the pixels where at the edge of the object were given by exponen...