In order to identify the panic crowd behavior with a more accurate and effective method, a new scheme is proposed which can utilize the intersection density of motion vector in the video to judge the abnormal panic crowd behavior. This algorithm is based on LK optical flow to extract information of motion vector from moving people, and to obtain the intersection between two motion vectors, then uses divided image blocks to get the intersection density which is the key to identify abnormal crowd. Experiments on several datasets show that this algorithm can identify the panic crowd behavior with high accuracy.