A Security-Aware and Environmentally Robust Hybrid YOLO-E–YOLOv12 Framework for UAV-Based Helmet Detection and Color Classification
Year: 2026
Type: conference-paper
Abstract: UAV-based automated monitoring experiences issues related to security because of environmental aberrations, undetected object categories, and inference instability. Safety in construction and industry requires helmets. This research employs the secure and robust hybrid detection framework (SRHDF), utilizing YOLO-E, an open-vocabulary vision-language model, alongside YOLOv12, a high-performance closed-set detector. The SAHI and illumination-invariant preprocessing are employed to ensure reliable and robust performance. The proposed SRHDF is tested on a balanced public PPE dataset and a new UAV helmet dataset with 6,471 annotated cases across altitudes, perspectives, lighting conditions, and simulated weather perturbations. YOLO-E accurately generalizes unseen helmet colors in zero-shot situations. The proposed SRHDF outperforms the state-of-the-art YOLO-12 and YOLO-E methods.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201880679