Motion Anomaly Detection in Surveillance Videos Using Spatial and Temporal Features

The problem of anomaly detection is defined as the task of detecting a deviation from usual conformity in a video sequence due to an event, and if such an event is found, determining its start and endpoints. Since anomalies seldom occur in the physical world, most datasets available for anomaly detection consist of only normal activities. In this paper, novel computer vision algorithms using spatial and temporal features to improve detection accuracy is proposed. Further optimizations would be performed by using principal component analysis to reduce the feature dimension. Finally, we would compare our results with existing literature on the Chinese University of Hong kong Avenue (CUHK Avenue) dataset.