KALABALIK GÖZETLEME ORTAMLARINDA ANOMALİ TESPİTİ
Abstract
Camera survaillance systems are effective security methods with a wide range of uses. Videos obtained from these systems are examined by the security personnel in order to determine the dangerous situations and take the necessary precautions. Increasing technological developments in recent years have led to reductions in the cost of cameras and an increase in the use of surveillance systems and the amount of video data being acquired. Processing these data manually is very hard and time consuming. The visual attention module of the human brain is limited and thus, human attention shows a great decline after a certain period of time. This is the serious problem in manual analysis of large amounts of data. Intelligent video surveillance systems reduce the need for human power and enable to obtain meaningful information from large amount of video data.
One of the important purpose of intelligent video surveillance systems is to analyse videos effectively to distinguish between normal and abnormal conditions and to alert the relevant operator about abnormal events. Although various methods are used to design intelligent surveillance systems, general approach is modeling normal events and identifying abnormal situations that do not fit into the model. The reasons for this approach are that the anomaly definition varies according to the content, namely, situations considered abnormal for a particular scene may be considered normal in another scene and the difficulties in finding the abnormal training samples.
In this study, multi-scale histogram of optical flow features (MHOF) and log-Euclidean covariance matrices are used in automatic anomaly detection with single class classification methods. Log-Euclidean covariance matrices are used for the first time to detect anomalies. Unlike traditional methods, which utilize gradient-based or optical flow-based features for motion representation, two important types of features that encode motion and appearance cues are combined with the help of covariance matrix. Covariance matrices are symmetric positive definite (SPD) matrices which form a special model of the Riemannian manifold and are not suitable for traditional Euclidean operations. Most of the computer vision algorithms are developed for data points located in Euclidean space. For this reason, covariance matrices are mapped to Euclidean space by utilizing log-Euclidean framework.
The model building process, which is the first step in the detection of abnormal situations, is performed by using features obtained from normal events with single class classification methods (Support Vector Machines, Support Vector Data Description). In the detection process, dissimilar events meaning that do not fit the model are marked as abnormal.
Experiments carried out on an anomaly detection benchmark dataset and comparisons made with previous studies within the scope of the study show that successful results are obtained in detecting abnormal situations.
Keywords: Anomaly detection, multi-scale histogram of optical flow, log-Euclidean covariance matrices, crowd motion analysis, one class classification.