针对物体基视频编码与分割应用，提出了求取特征区域真实运动矢量的一种算法。首先，利用运动检测器将图像序列的每一帧分割成静止区域与运动区域。然后只对运动区域进行运动估计。为了克服传统块匹配法的固有缺点，本文算法采用变尺寸块匹配法，大的均匀区用大方块，运动物体的边缘区用小方块；在进行块匹配搜索期间，在易误匹配的区域，为了不遗漏真实的运动矢量，先按照ＭＡＥ准则预选出多个候选运动矢量，尽可能保证真实的运动矢量包含其中，接着再根据相邻块运动矢量的相关性，从多个候选运动矢量中选出一个比较可靠的运动矢量。实验结果表明，该算法产生的运动矢量真实可靠，而且一致性好，可用于识别可靠的运动区域。 Aiming at the application of object-based video coding and segmentation, an algorithm for calculating the real motion vector in the feature region is proposed. First, each frame of the image sequence is divided into a still area and a motion area using a motion detector. Then only exercise area motion estimation. In order to overcome the inherent shortcomings of the traditional block matching method, this algorithm uses the variable size block matching method, the large uniform area with large square, the moving object edge area with small square; During the block matching search, In order to avoid omitting the real motion vector, a plurality of candidate motion vectors are pre-selected according to the MAE criterion to ensure that the real motion vectors are included as much as possible, and then the candidate motion vectors are selected according to the correlation of the motion vectors of adjacent blocks A more reliable motion vector. Experimental results show that the proposed motion vector is reliable and consistent, and can be used to identify reliable motion region.