Vision Based Drone-Human Differentiation for Aerial Security System

Abstract

The increasing use of unmanned aerial vehicles (UAVs) has raised serious security concerns, particularly in distinguishing unauthorized drones from humans in restricted areas. This study proposes a real-time vision-based detection framework using a lightweight YOLO11s model to perform accurate drone and human detection from video streams. A hybrid dataset consisting of 2,082 original images was expanded to 5,440 images through structured data augmentation to enhance model robustness and generalization capability. The trained model was evaluated through offline video testing and real-time deployment scenarios. Experimental results demonstrate that the proposed system achieves a precision of 0.912, recall of 0.903, and an F1 score of 0.907, indicating balanced and reliable detection performance. Furthermore, real-time experiments on heterogeneous hardware platforms show that GPU acceleration significantly improves system efficiency, achieving 12.4 frames per second (FPS) with an inference time of 81.0 ms, compared to 5.2 FPS and 191.8 ms on CPU. These findings confirm the effectiveness and practicality of the proposed framework for real-world aerial surveillance and security applications.