A contextual weapons authorization system using dual YOLOv8 detection for border security

This work presents a novel two-stage object detection system mainly designed for security applications in border and conflict areas. Using two YOLOv8 models, the system uses a cascading structure whereby one of the models detects firearms and activates a secondary model identifying army uniforms when detected. By means of Intersect over Union (IoU) computations, the system examines spatial relationships between identified weapons and uniforms, thus facilitating differentiation between authorized military personnel and possibly illegal individuals carrying weapons. When unauthorized weapon possession is discovered, the system creates real-time alarms with relevant detection information for security monitoring systems. With both YOLOv8-nano and YOLOv8-medium variants for every detection job, the architecture provides deployment flexibility, allowing for either complete processing in the cloud for maximum accuracy or edge computing implementation in remote locations. By lowering computational overhead and raising detection accuracy in settings where security is crucial, the suggested design model shows notable benefits over current methods.