Anomaly Detection In Compressed Video

In this paper, an anomaly detection approach has been developed on video compressed in H.265 format. In order to detect anomalies, the motion vectors in the compressed video and the region information of the motion vectors were used. This information was provided as input to the autoencoder model, which is an unsupervised artificial neural network method, and thus the model was trained. The trained model was tested on video data containing anomalies. As output, during the streaming of any video, it is provided to draw a regularity score graph and display the anomaly regions by color. In this paper, we propose an autoencoder based method for anomaly detection in compressed video instead of the original uncompressed video.