粒子滤波在非线性和非高斯问题上具有独特的优越性,但在视频跟踪过程中，其跟踪性能却在很大程度上依赖于观测模型的选择。为了解决被跟踪目标特征状态随时间变化而与粒子观测模型不匹配的问题，提出了一种新的粒子滤波算法，即将被跟踪目标的不同特征状态与粒子观测模型相结合，形成一组具有不同观测模型的粒子，并且在跟踪过程中，对应不同观测模型的粒子根据被跟踪目标所表现的特征线索的变化而相互转换，从而动态刻画了被跟踪目标特征变化的过程。实验结果表明，本算法能够有效处理由于头部旋转而导致跟踪性能下降甚至丢失跟踪目标的问题，提高了跟踪的准确性，并且具有较好的鲁棒性。

Particle filtering has drawn more attention recently due to its superior performance in nonlinear and non-Gaussian problems, which uses an observation model to describe the interested target and its performance depends strongly on the observation model. A novel particle filter algorithm has been proposed for head tracking, which focuses on finding a solution to the problem that the feature of the tracked object does not always match the particle observation model with changing time. In brief, it unifies the difference of the tracked object features with the particle observation models, and thus it forms a group of particles with different models to track the target. Based on the changing displays in the feature cues of the head with the rotation, particles with different observation models convert alternately in the tracking process. To evaluate the novel head tracker performance, some real sequences is tested and some results are shown that the new tracker is robust to the rotation of head in a cluttered background and has better tracking precision than the standard particle filter.