Artificial intelligence assisted classification for early detection of potential threats for strategically important organizations

This study examines the effectiveness of surveillance cameras in the context of security vulnerabilities and societal threats and aims to improve these systems using deep learning techniques. A dataset consisting of 1254 images captured from surveillance cameras was created and expanded to 3762 images using data augmentation techniques. The dataset was annotated into five distinct categories: military, hospital, school, square, and mall. State-of-the-art object detection models, including YOLOv9, YOLO-NAS, RetinaNet, Faster-RCNN, and EfficientDet-D7, were trained using PyTorch and OpenCV. Among these models, YOLOv9 achieved the highest mAP50 score of 98.9%. The analyses demonstrate that surveillance cameras are effective in enhancing security levels across all categories. The study reveals that the placement angles, lighting conditions, and positioning of cameras significantly influence the performance of security systems. Additionally, object detection in low-light and varying weather conditions is emphasized as critically important for the effectiveness of security systems. The analyses conducted across five different categories highlight the potential of surveillance cameras to detect security threats, particularly in public spaces. The findings underscore the importance of surveillance cameras in protecting critical infrastructure and taking preventive measures against societal threats.