We use neural nets to generate descriptors of videos where motion is observed, and classify events as normal/abnormal.