With the increasingly prominent public safety issues and the rapid gro wth of video data traffic,intelligent video surveillance abnormal behavior dete ction has become a research hotspot in the direction of computer vision.Because the training effect of abnormal behavior detection under supervised learning is easily affected by the unbalanced distribution of video surveillance data set, this paper uses unsupervised learning method to obtain the effect of abnormal be havior detection with temporal and spatial resolution.An abnormal behavior dete ction method based on multi-manifold spectral clustering is proposed,which use s image feature points to construct time series information features,uses manifo ld learning to construct spatial information features,and finally uses spectral clustering to obtain video abnormal behavior detection results unsupervised.Ex perimental verification is based on the UCSD data set,the method in this paper has bette r detection performance and faster calculation speed.