Materials, Methods & Technologies, Volume 6, 2012

Abstract: By anomaly detection, we mean the detection of patterns in the video or images that do not conform to the expected behaviour. We can define anomaly as rare or infrequent behaviour compared to all other behaviours, which is also referred to as an outlier. However, detecting abnormalities in video is a challenging problem since the class of all irregular objects and behaviours is infinite and thus no (or by far not enough) abnormal training samples are available. To handle this problem we refer to the task of classification between two classes when there are no labels for an abnormal class in the training data. We propose a framework that is based on the bag of visual words model that applied to the set of images or frames to get the robust descriptors. To learn the normal activity patterns we use the extension of support vector machine algorithm to the case of unlabeled data. An experimental evaluation of the framework is conducted with a large publicly-available dataset of crowd scenes. We demonstrate that the proposed approach outperforms state-of-the-art anomaly detection algorithms.