基于彩色的概率跟踪及其快速实现

针对基于彩色的概率跟踪中每一时间步每一个像点都要计算彩色直方图导致计算代价大大增加这一瓶颈问题,提出了一种减少直方图搜索区域从而提高计算效率的方法。根据前一帧中跟踪器的输出,在当前帧中建立一个基准窗口,并将其分割成若干个相等的子窗口,分别计算出它们的彩色直方图。然后根据像点滤波中不同像点的搜索区域与基准窗口的相对关系,找出重叠部分,从而可以利用该部分已知的子窗口的直方图而无需重复计算,只需计算不同部分即可,实验结果表明,该方法使计算量减少了50%以上。

Color-based Probabilistic Tracking and Its Quick Realization

Color-based probabilistic tracking has been proved robust and versatile within a probabilistic framework,relying on the principle of color histogram distance.The use of particle filter in the structure allow people to better handle color clutter in the background,as well as complete occlusion of the tracked entities over a few frames.But the building of too many histograms at each time-step becomes the bottle-neck that increases the computational cost drastically.To address this problem,we present a new approach to calculate color histograms.By building fiducial research region and segmenting it into several sub-regions at time t+1 based on tracker output at time t ,we calculate color histograms of the sub-regions and save them.Then we find the same region out between fiducial research sub-regions and the research region of every particle so that the histogram of the part will not be accounted for again.Experiments show computational cost can be reduced over half.