Abnormal crowd behavior detection is a highlv focused research area of the intelligent monitoring.The paper mainly aims at studying how to extract spatiotemporal characteristics of dense crowd video,and how to improve the efficiency of anomaly detection.Combining with human visual system(HVS),we proposed a novel method based on the analysis of the quaternion Fourier transform which is a fusion of spatiotemporal characteristics in order to extract features of dense crowd scene.It is proved by the experiment that the proposed method can describe the dense crowd scene from different aspects and achieve good effect of anomaly detection as well.