随着视频编解码技术的发展，H．264已成为最主要的标准之一。为了能从H．264码流中准确有效地分割和提取出运动对象，提出了一种压缩域下的对象检测算法。该算法主要利用压缩域下对象的运动矢量信息，先对矢量进行中值滤波预处理，目的是为了减少运动估计算法和对象实际运动所产生的误差。然后利用基于熵的原理，建立运动对象在空间和时间上的一致性模型。在该模型基础上，采用最大熵方法自适应获得阈值，检测得到运动对象。实验结果证明，该算法可以获得比较好的检测结果。

With the development of video encoder and decoder technology, H. 264 has become one of the most important standards. In this paper, we proposed novel moving object detection based on spatial-temporal method in compressed domain, which mainly depended on the motion vector field and efficiently segmented the object. Firstly, a spatially median filter was applied to the motion vector to alleviate the error vector due to difference between the motion vector obtained from motion estimation and the real motion information. Secondly, we built the spatial and temporal coherence model with the entropy based method. We obtain the threshold with these models. The effectiveness of our method was demonstrated with its performance on experiment results.