Video Anomaly Detection Using Ensemble One-Class Classifiers

In this paper we present a novel algorithm for video anomaly detection. It is based on multiple local cells, which are acquired by splitting entire monitor scene. At each local cell, we group all feature vectors with clustering algorithm based on minimum spanning tree, and further model all groups using improved one-class SVM to build ensemble classifiers. For any new features at each local node in incoming video clips, we use the corresponding learned ensemble classifiers to estimate maximum abnormality degree. The proposed approach has been tested on publicly available datasets with frame-level and pixel-level criteria, and outperforms other state-of-the-art approaches.