A Comparative Study on Deep Learning based Semi-Supervised Video Anomaly Detection Methods
Webinaire organisé par le CeRVIM, animé par l’auxiliaire de recherche Mohammad Baradaran, associé au Laboratoire de vision et systèmes numériques. À cette occasion, M. Baradaran offrira une analyse des différentes solutions de détection supervisée d’anomalies vidéo soutenues par l’apprentissage profond.
Présentation de la conférence
Nowadays, high-quality cameras are ubiquitous and a huge amount of video data are being recorded for different purposes (such as surveillance in public places, traffic control on highways, border control, studying human or animal behavior, production quality control, etc.). Analyzing this huge amount of data is beyond the capability of human operators, hence there is a need for intelligent systems to analyze video content and to detect events of interest automatically.
Video anomaly detection (abnormal event detection) is one of the hot research topics in computer vision today, as abnormal events contain a large amount of information. Anomalies are the events that deviate from the majority of observed events and are one of the main detection targets in surveillance systems and most of the suspicious events which take place belong to this group. For example, a car moving with the speed of 120 km/h on a road (with a maximum accepted speed of 120 km/h), on a snowy day while other cars drive slowly, would constitute an abnormal event.
In this presentation we will critically analyze state-of-the-art deep learning based semi-supervised video anomaly detection approaches, analyzing the strategies and pointing out their strong and weak points. These results are used and presented in our review paper. Moreover, the results of these experiments show the existing shortcomings in the field and provided the basis for our proposed method. Our proposed method will be presented in a forthcoming presentation.
Restons en contact!
Vous souhaitez être informé des nouvelles et activités de l'IID? Abonnez-vous dès maintenant à notre infolettre mensuelle.