RZUD: A Novel Hybrid Model for Small Sized Handgun Detection

Abstract

Closed-circuit television (CCTV) cameras have become ubiquitous tools for security, supplemented by an active system that can automatically detect firearms, a measure intended to discourage criminal activities like gun violence. However, accurately identifying small handguns poses a unique challenge due to their lack of distinguishing features. This deficiency leads many existing algorithms to produce false positives and negatives. To address this issue, a novel hybrid model named RZUD (RoI-ZOOM-UNBLUR-DETECT) has been developed. RZUD operates in four stages: selecting regions of interest, zooming in on selected regions, unblurring the resized regions, and ultimately performing detection. This comprehensive approach significantly improves detection accuracy. In empirical evaluations, RZUD outperformed state-of-the-art object detection algorithms including YOLOv3 and YOLOv7. When tested on a small-sized handgun dataset, YOLOv3 registered a 56% F1 score, but when combined with RZUD, this figure improved to 76%, marking a 20% improvement. Similarly, YOLOv7's F1 score rose from 56% to 77% when coupled with RZUD, a remarkable 21% gain. In essence, RZUD's novel methodology effectively elevates small handgun detection accuracy.