Towards a Usable Crime-Based Anomaly Detection Model

With an ever-increasing amount of video footage being gathered, the interest in automated anomaly detection in video has seen a large increase. This paper aims to go beyond the technical implementation of a video anomaly detection model and look into what is needed to take this technology into practice. By describing the legal and ethical aspects of automated anomaly detection in video and gathering feedback from the Netherlands Forensic Institute, a novel anomaly detection methodology is created and tested. The proposed methodology uses I3D features to provide a class-agnostic method of detecting anomalies in video. Although performance is not at the level of the current state of the art, the agnostic nature as well as the analysis and discussion of relevant literature, provide an important step to bring these technologies into practice.