Real-Time Weapon Detection based on Optimized YOLOv8

Abstract

Shootings in schools, shopping malls, and other public places are increasing every year, partly due to the ease of legally acquiring firearms in some nations. This research proposes an automated weapons detection system utilizing optimized transfer learning with the YOLOv8 architecture to analyze CCTV footage in public spaces. The study uses a modified YOLOv8 model with pre-trained weight for real-time object identification to improve security response capabilities. Experiments reveal that the optimized YOLOv8m configuration has a higher detection accuracy and processing efficiency. The model achieved an F1 score of 91.9%, a mean average precision of 94.2%, and a processing speed of 35.7 frames per second using Stochastic Gradient Descent with a momentum of 0.9 and a learning rate of 0.001. The findings show that the proposed approach identifies weapons successfully in real-world public situations, proving its potential to considerably improve public security.