Dense spatio-temporal features for non-parametric anomaly detection and localization

In this paper we propose dense spatio-temporal features to capture scene dynamic statistics together with appearance, in video surveillance applications. These features are exploited in a real-time anomaly detection system. Anomaly detection is performed using a non-parametric modelling, evaluating directly local descriptor statistics, and an unsupervised or semi-supervised approach. A method to update scene statistics, to cope with scene changes that typically happen in real world settings, is also provided. The proposed method is tested on publicly available datasets and compared to other state-of-the-art approaches.