中文摘要：

针对非刚体目标的精确实时跟踪问题，提出了一种融合先验形状信息的基于最稳定极值区域（MSER）检测器的跟踪算法。首先，利用讥练样本建立目标颜色特征的混合模型，生成目标统计颜色概率图，为最大稳定区域方法提供概率统计依据。其次，利用基于最稳定极值区域方法给出最稳定的分割结果。最后，利用训练样本得到目标的先验动态形状模型，并且融合目标形状信息与通过MSER算法生成的稳定区域信息，去除虚假分割结果，提高目标检测精度与跟踪性能。实验结果证明，该算法能在视频序列图像中有效检测并跟踪目标。

英文摘要：

An exact non-grid object detecting and tracking algorithm is proposed which combines the Maximally Stable Extremal Region（MSER） with the object shape prior knowledge. The first step of the algorithm is the calculation of multivariate Gaussians of color likelihoods which will be passed to the MSER. Secondly, based on analysis of MSER, the maximally stable boundaries are exploited by finding the connected regions of interest. Finally, by obtaining the object shape prior model from training set using Principal Component Analysis（PCA） and combining the shape prior information and the MSER information, an active weight model is provided in the accurate object segmentations in each frame and robust multiple object tracking. The validity of the method is proved robust by experimental evaluations.