vu-videoanomaly-2019.pdf (16.39 MB)
Video anomaly detection using deep generative models
thesisposted on 2019-11-30, 00:00 authored by Hung Thanh Vu
Video anomaly detection faces three challenges: a) no explicit definition of abnormality; b) scarce labelled data and c) dependence on hand-crafted features. This thesis introduces novel detection systems using unsupervised generative models, which can address the first two challenges. By working directly on raw pixels, they also bypass the last.