Reducing Gun-Related Incidents on Construction Sites: An AI-Driven Approach for Automated Detection of People, Tools, and Firearms

Traditional construction safety research has primarily focused on occupational hazards. However, workplace violence, particularly firearm-related threats, remains a critically underexplored issue in the field. High-profile incidents such as the tragic shooting on September 2024 in Phoenix, Arizona, where a 48-year-old construction worker was killed and a 19-year-old coworker seriously injured, highlight the urgent need for violence mitigation strategies on construction sites. To address this gap, the present study introduces a novel dataset that integrates the YouTube Gun Detection Dataset (YouTube GDD) and the YOLO7 Power Tool Dataset. This dataset supports multi-class detection of people, firearms, and tools, effectively distinguishing between visually similar objects such as drills and guns. The study evaluates multiple YOLO model configurations, with YOLOv11x achieving optimal results of 92.0% precision and 91.2% mAP at 0.5. These findings underscore the promise of object detection technologies in proactively preventing workplace violence, thereby enhancing construction site safety.