提出了一种红外图像中遮挡人体目标的分割方法。首先通过传统的阈值选取方法或直方图聚类等红外图像分割的方法，获取遮挡区域目标的二值化模板。用目标像素在模板中的相对行列坐标作为特征集使用Mean Shift算法分别计算各像素在行列方向的收敛位置并使用复数向量进行联合表达，再次以所有的复数向量作为特征集进行Mean Shift聚类，根据各像素位置对应的复数向量所属类别对其进行划分，完成遮挡目标的分割。与分水岭算法相比，该算法的分割结果完整保留了目标模板的外形，并且可以通过Mean Shift 带宽参数的选择完成不同精细程度的分割。

An occluded human segmentation method in infrared image was presented in this paper. Firstly, the occluded regions binary mask was obtained using traditional infrared image segmentation methods such as threshold selection or histogram clustering. Then, two mean shift clustering process were carried out separately using the objects relative row and column coordinates to the mask as feature sets. Each pixels convergence position in vertical and horizontal direction was represented by a complex vector. Finally, another mean shift process was carried out on all the complex vectors, and the object masks pixels were classified according to their corresponding complex vectors cluster. Thus, the objects were segmented in the binary mask. Compared with the watershed algorithm, the segmented results of our algorithm were complete and could be controlled from coarse to fine by changing the mean shift bandwidth parameters.