基于自适应分块外观模型的视觉跟踪

Fragment-based visual tracking with adaptive appearance model

提出将目标和遮挡物视为关联对象,利用遮挡物的特征来更新模板,引入高斯混合模型(GMM)进行自适应的模板学习。模板中的每个像素利用包含3个分量的GMM来表示。在当前帧中,一旦获得模板匹配,模板中每个像素及其对应的GMM都会进行相应的更新。实现结果表明,对于部分遮挡和外观变化,算法能够实现顽健的跟踪。

A joint object was proposed which combine the target and the occluder in order to exploit the characteristics of the occluder to evolve the target template.To this end,the Gaussians mixture model（GMM） was introduced to enable an adaptive learning mechanism.Each pixel of the template is modeled using a GMM with three components.Once the template-matching is achieved in the current frame,the pixel value is used to update a specific component in its GMM.After that,the pixel will also be updated.Experimental results show that the proposed method is robust with respect to the partial occlusions and natural appearance changes.