Unsupervised video anomaly detection using feature clustering

This study addresses the problem of automatic anomaly detection for surveillance applications. A general framework for anomalous event detection in uncrowded scenes has been developed which consists of the following key components: (i) an efficient foreground detection model based on a Gaussian mixture model (GMM), which can selectively update pixel information in each image region; (ii) an adaptive foreground object tracker that combines the merits of Kalman, mean-shift and particle filtering; (iii) a feature clustering algorithm, which can automatically choose the optimal number of clusters in the training data for scene pattern modelling; (iv) a statistical scene modeller based on Bayesian theory and GMM, which combines trajectory-based and region-based information for enhanced anomaly detection. The resulting approach achieves fully unsupervised anomaly detection in surveillance video. The experimental results show improved detection performance compared with the state-of-the-art methods.