Abstract

In this paper we address the abnormality detection problem in crowded scenes.We propose to use Generative Adversarial Nets (GANs), which are trained usingnormal frames and corresponding optical-flow images in order to learn aninternal representation of the scene normality. Since our GANs are trained withonly normal data, they are not able to generate abnormal events. At testingtime the real data are compared with both the appearance and the motionrepresentations reconstructed by our GANs and abnormal areas are detected bycomputing local differences. Experimental results on challenging abnormalitydetection datasets show the superiority of the proposed method compared to thestate of the art in both frame-level and pixel-level abnormality detectiontasks.