Authors: M. Pradeepthi, A. Shravani
Abstract: As the number of crimes committed with firearms and other sharp weapons continues to rise, there is an urgent need for an intelligent monitoring system capable of automatically detecting potential threats. It is not simple to spot knives and guns in surveillance images due to their small size, unpredictable orientations, and complex backgrounds. This research presents a method for the automatic detection of weapons, such as knives and guns, by use of cutting-edge deep learning object recognition algorithms. As its primary detection model, this system compares YOLOv7 against other variations of YOLO, including YOLOv7-x, YOLOv7-w6, YOLOv7-e6, and YOLOv8n. The detection models are evaluated and trained using a tagged dataset consisting of photographs of firearms and knives. The bounding box annotations allow for precise localisation of the weapons inside the input pictures. To assess how well each model performs, metrics including Precision, Recall, mAP@0.5, and mAP@0.5:0.95 are used. Experimental analysis found that the YOLOv7-e6 model had the highest mean Average Precision of the models evaluated, indicating improved detection capabilities and the capacity to more consistently identify weapons like knives and guns in surveillance circumstances. Whether for civilian or military usage, the proposed architecture streamlines the process of intelligent security monitoring and enhances automatic detection of potentially dangerous weapons.
DOI: http://doi.org/0.5281/zenodo.21374697
