Authors: Narkhede Chaitali Hemraj, Prof. J. M. Patade, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Public places such as railway stations, airports, schools, malls and religious gatherings need to detect weapons quickly to prevent violent incidents. Human operators monitoring many camera feeds or reading X-ray baggage images can miss threats because of fatigue and the rarity of real events. This paper reviews AI-based weapon detection in two settings: visible-light CCTV video, where handguns, rifles and knives must be detected in real time, and X-ray security screening, where prohibited items must be recognized in cluttered, overlapping baggage images. It examines deep object detectors, including two-stage region-based networks, single-stage YOLO-family detectors, feature pyramid networks and focal loss for small and rare objects, and surveys specialized datasets and studies for handgun detection in video and threat detection in X-ray imagery. Practical issues such as small object size, occlusion, similar-looking objects, low-light video and false alarms are analysed. A five-layer architecture comprising video and X-ray input, preprocessing, a detection engine, alert verification by human operators and a response layer is proposed, with research gaps, objectives, a comparison of detectors and challenges including privacy and bias. The review concludes that AI can strengthen security when it acts as an alert system supervised by trained personnel.

DOI: https://doi.org/10.5281/zenodo.24218949