一种基于图模型的粒子滤波跟踪方法

针对复杂背景下目标被部分遮挡时的稳定跟踪问题，提出一种基于图模型的粒子滤波跟踪方法。该方法将图模型应用于粒子滤波之中。首先，融合颜色和边缘特征建立目标的观测模型，构建粒子滤波框架。然后，选取目标特征区域，将一个目标分成几个部分，每一部分作为图的一个顶点，建立图模型。最后，将图模型应用于粒子滤波，目标跟踪过程中，图模型中每一个部分的状态信息可以传送给其它部分。实验结果表明，当目标被部分遮挡的情况下，该方法能够估计出遮挡部分的状态，实现稳定的跟踪目标。

Particle Filter Tracking Algorithm Based on Graphical Model

In complicated scene, in order to tackle to track stably objects whose part were temporally occluded, the paper presented a method for particle filter tracking algorithm based on graphical model. It incorporated graphical models into particle filtering. Firstly, by the fusion of color and edge character, set up target observation model, and create particle filter framework. Then, select target feature area, this method used not one region of the whole of the object but the multi-part region of it if it was divided in some parts. Create graphical model by using the multi-part region. Finally, incorporate graphical models into particle filtering, in the process of target tracking, messages about the state of the parts were sent to other parts in the graphical model. On the experiments of tracking people, as the result, when the tracked object has occluded part, the proposed method can infer the state of the occluded part, and track stably.