| Sumario: | Surveillance camera is one of the foremost tool for monitoring movements of human and for preventing unwanted and unintended activities. Video Surveillance demands crime and avert unfortunate consequences which impacts human society. Convolutional neural network has shown promising future object detection and recognition, particularly in images and videos. However, labels are required for learning as convolutional neural network is a supervised technique. For anomaly detection in videos, we propose a spatiotemporal architecture which includes representation of spatial feature and evolution of spatial features. Benchmark confirms the accuracy of our proposal in comparison with state-of-the-art method. Hence, abnormal and potentially unsafe activities which are even hidden from a security personnel will be easily detected. The key motivation for development and deployment of our proposed method is to avoid existing human error while monitoring and to ensure security of a person or organization's security. It is well known that anomalies are highly contextual. For example, riding a bicycle in park or road would be normal but riding the bicycle in pedestrian path would be an anomaly.
|
|---|