Lightweight Smart Helmet Detection in Smart Construction Sites Supported by Unmanned Aerial Vehicle Networks

ABSTRACT Conventional deep learning–based helmet detection on construction sites is constrained by fixed camera perspectives, high computational demands, and challenges in multi‐scale detection. Moreover, transmitting video data to the server seriously hinders real‐time and efficient monitoring. To address these issues, this paper proposes an integrated smart detection system that combines unmanned aerial vehicles (UAVs) and 6G‐enabled Internet of Things (IoT) technology with an enhanced real‐time detection Transformer (RT‐DETR) framework. By embedding graph convolutional operations in the RT‐DETR model, it effectively captures both local and global dependencies, thereby improving feature representation for multi‐scale helmet detection. UAVs provide adaptive, wide‐area visual coverage, while 6G IoT supports ultra‐reliable, low‐latency communication for real‐time data transmission and system coordination. Experiments demonstrate that our approach achieves higher accuracy and robustness in complex environments while maintaining computational efficiency suitable for real‐time deployment in next‐generation smart construction systems.