Integrated Intelligent Surveillance System Using Deep Learning

Abstract

Nowadays the surveillance systems are widely used to find out the suspicious events that have occurred. In conventional systems, there are a lot of limitations such as storage, bandwidth, cost, the short lifespan of hardware devices, loading issues, etc. We developed an intelligent surveillance system using deep learning in which the video footage of suspicious events is extracted. Transfer learning, a part of machine learning, is used for face detection which involves the reuse of a pre-trained model on new data. The abnormal activity detection is done using a multi person MoveNet Light model and the face detection is done using VGG16. The suspicious objects found in the frame (gun, mask) are identified using corner detection. This system offers less bandwidth, high security, effective storage, and reduced load-balancing issues. In this paper, we detailed the face detection, object detection and anomaly detection used in our system.