CAJ | 학술논문

针对原始的Mean Shift跟踪算法虽能准确地估计目标位置，但对目标尺度和方向不能实现自适应估计，结合目标模型与候选目标区域的候选模型得到了反向投影图，此反向投影图可表示图像中像素点属于目标的概率，将反向投影图的矩特征应用到原始Mean Shift跟踪算法框架，实现了目标尺度和方向适应性Mean Shift跟踪。实验结果表明：该算法能有效跟踪尺度和方向变化的目标。
침대원시적Mean Shift근종산법수능준학지고계목표위치，단대목표척도화방향불능실현자괄응고계，결합목표모형여후선목표구역적후선모형득도료반향투영도，차반향투영도가표시도상중상소점속우목표적개솔，장반향투영도적구특정응용도원시Mean Shift근종산법광가，실현료목표척도화방향괄응성Mean Shift근종。실험결과표명：해산법능유효근종척도화방향변화적목표。
Although the position of target can be well obtained by the original mean shift tracking algorithm, the target’ s scale and orientation can not be estimated adaptively.This paper proposed an improved mean shift algorithm to address the issue of how to estimate target’ s variation of scale and orientation.Firstly, by combining the target mode and the candidate model, we obtain the weight image which represents the possibility that a pixel belongs to the target.Then the moment features of the weight image are applied in the mean shift tracking framework, so that the scale and orientation changes of the target can be adaptively estimated.The experimental results demonstrate the effectiveness of the proposed algorithm when tracking the target with changes in scale and orientation.