Incident-Aware Geofenced UAV Surveillance with YOLO-Based Object Detection for Industrial Facility Security

We design and evaluate an incident-aware, geofenced UAV surveillance system for an expanding industrial yard. The architecture integrates GPS-enabled multirotor drones, real-time object detection (YOLOv11-nano), and a geofencing stack that combines signed-distance safety margins, a quadratic programmed control barrier function (CBF) filter, and model-predictive control (MPC) for constraint satisfaction under time-varying hazards. Using VisDrone2019 with a reproducible 70/15/15 split, the detector attains mAP@0.5 = 0.912; an operating threshold τ = 0.185 yields the best F1 (≈ 0.61), balancing precision and recall for dense yard scenes. We produce exportable patrol artifacts (CSV/GeoJSON) to support deployment and auditing. Simulation studies show enforcement of safety margins without excessive path inflation, and multi-UAV sectorization enables deconflicted coverage with incident-aware replanning. All training scripts and route files are released for reproducibility. Subject to on-site validation of latency and false-alarm rates, the system is ready for controlled pilot deployment at TLG—Denton and serves as a model for scalable, regulation-aligned UAV security operations.