In terms of the problem that the traditional detection method of crowd abnormal events based on group motion state analysis does not describe the semantic information of scene adequately, the Girvan-Newman (GN) splitting up algorithm found by the community in the complex network is introduced. The pedestrians with similar motion characteristics and similar positions are divided into multiple groups, and the differences among the groups in normal and abnormal scenes are described and the occurrence of abnormal events is detected with the changes in group motion intensity and group number. Through experimental verification, the proposed algorithm can accurately detect abnormal events while enriching the semantic information of the scene.