Detecting Anomalies in Surveillance Videos with Spatio-Temporal Features
AbstractOne of the purposes of video surveillance systems is to detect anomalies which are unexpected situations at a certain location or at a frame. Anomalies can be related to motion or appearance according to its spatial position. In this paper, we propose an anomaly detection system based on spatio-temporal features. Features from Accelerated Segment Test (FAST) is used for detection of corners location. Optical Flow magnitude and orientation of these points is used as spatio-temporal features. A grid is to the frames to neutralize the effect of proximity to the camera. Normal patterns are clustered with an unsupervised neural network so called Self-Organizing Maps (SOM). In test videos if extracted features cannot model with normal clusters, associated grid cell will be marked as anomaly Keywords - Video Surveillance, Anomaly Detection, Features from Accelerated Segment Test (FAST), Optical Flow, SelfOrganizing Maps (SOM)