基于多测量融合的粒子滤波跟踪算法

复杂背景下运动目标的可靠跟踪,是计算机视觉领域中一个极具挑战性的问题。提出了一种融合颜色和纹理信息的粒子滤波跟踪算法,在粒子滤波的测量阶段,使用颜色直方图对目标进行颜色描述,用梯度方向向量对目标进行纹理描述。对这两种信息,分别用Bhattacharyya系数和欧几里德距离比较粒子与参考模板的相似性。为解决目标变化和遮挡问题,采用了模板更新策略。实验结果表明该方法是稳健的,能够在复杂的背景下对运动目标进行有效、可靠的跟踪。

Multi-measurement Fusion for Visual Tracking by Particle Filter

Visual tracking in a clutter background remains to be a challenging task by far.The particle filter based tracking algorithm proposed in this paper fuses color and texture information to build a robust measurement function.During the measurement step,the color information and texture information were represented by color histograms and gradient orientation vector respectively.Bhattacharyya coefficient and Euclidean distance were used to set up an effective connection between the estimated model parameters and the image likelihoods.Moreover,to overcome the problem of appearance changes,partial occlusions and significant clutter,an adaptive model update method was adopted.Experimental results show that the proposed method is robust and effective.