At present, the video background extraction algorithm of static scene has been nearly mature. However, video background extraction in dynamic scenes remains a challenge. In order to solve this problem, this paper proposes a dynamic scenes video background extraction algorithm. Here, our dynamic scene is based on camera movement. Firstly, we detect saliency target of video frame according to context information and do a processing of fuzzy enhancement. Meanwhile, we analyze flow filed by SIFT Flow method to do a nonlinear fusion with fuzzy enhancement result. Up to now, we can obtain moving target .Because of other gray information in the foreground affect target extraction, so we have to do a process of binarization and find out bounding box. After these preparations, we will track moving object with real-time algorithm. Finally, we use KNN algorithm to get accurate moving targets. Experiment results show that the proposed method for dynamic scenes video background extraction could get better results.