Identifying Abnormal Map in Crowd Scenes Using Spatio-Temporal Social Force Model.

Abnormal map defines the distribution of potentially dangerous in the crowd scenes. In contrast to other crowd behaviors, it has received a lot of attention among the video surveillance community to prevent abnormal situations. In this study, we conceive the abnormal map according to interaction forces of individuals. For this purpose, a grid of particles is first placed over the image and it is adverted with the space-time average of optical flow. By treating the moving particles as individuals, their change of interaction forces are then estimated using spatio-temporal Social Force Model (st-SFM). The st-SFM jointly learns and combines crowd density, intra-group stability and inter-group conflict for computing interaction forces. Finally, the abnormal map is obtained by mapping the change of interaction forces into the image plane. The experiments are conducted on a publicly available dataset from University of Minnesota for escape panic scenarios and a challenging dataset of crowd videos taken from the railway stations. The experimental results have demonstrated that such an abnormal map is not only useful but also necessary for abnormal crowd analysis and understanding.