Real-time Gun Detection using YOLO: A Deep Learning Approach for Public Safety

Gun violence is a pressing issue worldwide, necessitating the development of robust and efficient systems for firearm detection in real-world scenarios. This research extensively evaluates state-of-the-art object detection models for real-time firearm detection. We investigated a real-time gun detection systems leveraging the You Only Look Once (YOLO) object detection framework to find the best gun detection architecture. Our proposed approach uses the YOLO algorithm with ResNet backend (by replacing ELAN) to detect firearms in real-time security and surveillance applications, eliminating the need for separate models on different devices by focusing on single-device implementation. It is trained on an extensive dataset of annotated firearm images in different environments and contexts. To address the challenges associated with firearm detection, we employ different YOLO model versions (YOLO version 5s, 7x, 8s, and 9c) to ensure the model’s robustness and generalization capability. Experimental evaluations demonstrate the efficacy of YOLOv8s in accurately detecting firearms in real-world settings while maintaining high throughput, with inference speeds exceeding 95 frames per second on a standard GPU. For extensive analysis, we experimented with the older YOLOv5s and YOLOv7x models and the newer YOLOv9c model where YOLOv5s outperformed the newer YOLOv7x and YOLOv9c models in term of detection accuracy. However, the YOLOv5s takes a longer frame processing time.