Wasserstein clustering based video anomaly detection for traffic surveillance

Abstract

Anomaly Detection is very important in present scenario with huge availability of data and enormous difficulty in extraction of meaningful information out of it. In this paper we present an approach for video anomaly detection based on trajectory features and spatio – temporal features. Clustering of spatio – temporal features and trajectory features are performed in Wasserstein metric space and cluster distance and span in Wasserstein metric space is exploited to perform anomaly detection. The Performance of the Anomaly detection with Wasserstein distance based K – means and Wasserstein distance based DBSCAN clustering of the 3D wavelet features and trajectory features was studied. The method is robust and suffers from fewer false alarms.