Automated AI Surveillance System for Real Time Person and Weapon Detection

Abstract

Having safety and security in both the public and private space requires real-time and precise surveillance, since within a very short time, not disclosing the threat, the threat will grow to an extreme scenario like armed assault, robbery, vandalism or mass-hysteria. The conventional surveillance systems rely on intensive human surveillance that is time-consuming, intermittent and falls under the subject of fatigue-related errors. To address these issues, this paper introduces an automated AI-based surveillance system that can identify real-time danger through the use of computer vision and deep learning systems. First, video streams are post-processed to improve the video quality and clarity using motion detectors, frame difference filters, and noise cutters. Segmentation techniques are then employed to localize the regions of interest to eliminate unwanted information in the background. Deep learning algorithms like YOLOv8, SSD, and Faster R-CNN will be used to detect people, weapons, vehicles, fire, and suspicious persons. The features based on spatial, motion and context of the identified regions are used to classify the types and degree of threat. The results of experiments on the COCO dataset and personal weapon dataset prove that YOLOv8 is more accurate and real-time. The findings show that the suggested system is advantageous in terms of detecting reliability, minimising the reliance on human surveillance, and responding quicker to security risks.