Design and Performance Evaluation of a Real-Time Computer VisionBased Weapon Detection System for School Surveillance in Philippine Educational Institutions

Abstract

School violence involving firearms and edged weapons has become a pressing concern in Philippine education, with the Department of Education reporting that only about 36% of the country's roughly 48,000 public schools have CCTV cameras, and existing manual security procedures remain inconsistent and prone to human fatigue. Most published weapon-detection systems, meanwhile, are designed and tested only on expensive, surveillance- grade hardware, which limits their applicability in underresourced Philippine schools. This study designed and evaluated a real-time computer vision-based weapon detection system built around consumer-grade cameras (webcams and phone cameras) rather than specialized surveillance equipment. A YOLOv11s model was fine-tuned on a custom dataset of 30,525 images covering six weapon classes: knife, handgun, rifle, grenade, shotgun, and brass knuckles. Detection performance was evaluated using accuracy, precision, recall, F1-score, false positive rate, false negative rate, and mean confidence score per class, while detection latency was measured through live testing. Per-class accuracy ranged from 93.27% to 99.03%, with strong precision and recall for grenade, handgun, knife, and brass knuckles, though rifle and shotgun detection showed comparatively weaker recall and precision. The system achieved an average detection latency of 2.149 seconds. These results indicate that a low-cost, camera-based weapon detection system can meaningfully extend automated security coverage to schools lacking CCTV infrastructure or dedicated security personnel.