An Analysis of Abnormal Event Detection and Person Identification from Surveillance Cameras using Motion Vectors with Deep Learning

In many places, like museums, banks, and airports, network cameras have taken the place of outdated analogue cameras as a result of advancements in chip technology and the reduced costs of storage and bandwidth equipment, among other factors. In an endeavor to increase public safety and health protection, lower crime, and implement video surveillance, the sector has entered the "blowout" phase. A crucial part of developing an intelligent CCTV system is the detection of abnormal events, human behaviour, and object recognition. These technologies enable the detection of anomalous environmental phenomena, anomalous human behaviour, and the environmental state of alert. These systems use machine learning and machine vision capabilities to recognize and detect specific anomalies that appear in CCTV stream footage. For these systems, supervised learning is a popular training method, and frame-by-frame processing is frequently used. However, supervised learning has been replaced as unsupervised learning as well as semi-supervised learning for the system's training since anomalies come in a variety of forms and because it is not practical to which was before and educate all types of anomalies. By using this technology, the amount of labour that must be done by humans to manually spot anomalies in the CCTV live feed and generate alerts can be reduced or eliminated. Additionally, the method improves storage efficiency by only preserving anomalous occurrences in their original quality and leaving regular circumstances in low quality. Additionally, you may use the Grassmann algorithm to identify faces in CCTV data and alert surrounding security systems. The suggested approach offers better reliability in outlier detection and face detection, according to experimental findings.