针对相机运动引起的图像序列运动的问题，提出了一种基于聚类的相位相关块匹配运动估计算法。利用Harris算子分别在相邻帧图像上检测角点，以参考图像角点为中心选取一个矩形块，将块匹配法与相位相关相结合来计算图像间的运动矢量。最后，对获得的多个块的平移量，进行空间聚类从而选取运动估计比较准确的点。实验结果表明：该算法配准精度能达到亚像素，稳定性较好。

A motion estimation algorithm was proposed based on spatial clustering combined with the phase correlation block matching. The feature points in adjacent frames were detected respectively by using the Harris operator. The square block whose center was the feature was chosen. The motion vector was calculated by the block matching method and the phase correlation method. The points with the higher accuracy were picked up. The experiment results prove that the matching accuracy achieves sub-pixel level, and the stability is good.