We have developed an algorithm to detect anomalies in online surveillance video quickly and accurately. This technology incorporates two modules, the first of which is a novel feature extraction tool that generates realistic images and videos for future frame prediction. The second module includes a statistical algorithm that uses extracted features for anomaly detections. The techniques can be implemented in connection with hardware, software or both. It also provides a decision-making strategy with the desired false alarm rate which is crucial for minimizing human involvement.
Depiction of the System Process of Detecting Anomalies