隐马尔科夫测量场在区域跟踪中的应用

为了提高复杂背景下视频对象跟踪的有效性和鲁棒性，提出了一种基于隐马尔科夫（HMMF）模型与特征一空间联合分布的视频序列对象跟踪算法，其特点是将视频序列的对象跟踪问题描述为将当前帧分割为跟踪区域和非跟踪区域的动态图像分割问题．基于HMMF模型，将最优标记场估计问题转化为连续函数最优化问题．同时，将跟踪区域和背景区域的特征-空间联合概率分布作为贝叶斯估计中的条件分布，并且引入了改进的快速高斯变换来高效地估算分布函数．实验表明，这些技术使得跟踪算法对区域的局部形变和遮挡具有很高的鲁棒性，且具有较高的效率．

A New Region Tracking Method Based on HMMF Model

In order to improve the robust and efficiency of object tracking,a hidden Markov measure field(HMMF)-based tracking method was proposed.It describes the tracking procedure as a segmentation of current frame into a set of non-overlapping regions: tracking region and non-tracking region.Based on the key idea of using HMMF model,the optimal estimation for the label field is found in a function optimization framework.The feature-spatial probabilistic representation of a region as the conditional distribution was exploited in the Bayesian framework and the improved fast Gaussian transform(IFGT) method was adopted to further alleviate computational cost.Very promising experimental results on some real-world sequences were presented to illustrate that this tracker is robust to local deformation and partial occlusion.