Improving Handgun Detection: A Review and Proposal for Knowledge Graph Integration

Gun violence is a significant social issue, and in order to successfully identify firearms, advanced surveillance systems must be developed. Even with the introduction of deep learning algorithms like YOLO [5] and Faster R-CNN [61], it is still challenging to de-tect hidden weapons due to dynamic backdrops, shifting illumination, and partial object visibility. Even while current methods achieve high accuracy under ideal settings, they suffer significantly in real-world scenarios such object occlusion. Current handgun de-tection techniques are examined in this paper, which also divides them into deep learning and traditional approaches and identifies their drawbacks. We suggest combining knowl-edge graphs to tackle occlusion issues and false negatives by employing contextual and semantic links. Experimental validation demonstrates substantial improvementsin preci-sion (94.1%) and F1-score (92.6%) compared to standalone deep learning models, with a 59% reduction in false negatives for occluded objects. The purpose of this study is to stim-ulate additional developments in reliable and all-encompassing firearm detection systems for public safety.