Deep learning-based intelligent surveillance model for detection of anomalous activities from videos

For safeguarding and monitoring purposes, public places are equipped with surveillance cameras. Timely and accurate identification of suspicious activities is paramount to securing the public places. Assigning human personnel to keep continuous watch over ongoing activities is error-prone and laborious. To alleviate the need of human personnel for monitoring such videos, automated surveillance systems are required. This paper proposes a deep learning based intelligent surveillance model for detection of anomalous activities. The problem of anomaly detection has been handled as one class classification problem. The proposed approach involves two dimensional convolutional auto-encoder for feature learning, sequence-to-sequence long short term memory model for learning temporal statistical correlation and radial basis function as activation function in fully connected network for one class classification. We experimented on real-world dataset by two variants of proposed approach and achieved significant results at frame-level anomaly detection.