Real-Time Detection of the Environment Features in Infrastructure Construction Sites Based on Enhanced YOLOv8

Real-time detection of environmental features on infrastructure construction sites is crucial, especially for monitoring the safety equipment usage of construction workers. To adapt to different construction sites and improve detection efficiency, this paper designs a recognition method based on images collected by Unmanned Aerial Vehicles (UAVs). However, recognizing safety compliance in UAV-captured images presents significant challenges due to the small size of workers and complex backgrounds. To address these issues, this paper proposes a novel recognition framework based on YOLOv8, named YOLOv8Enhanced (YOLOv8-E). The proposed model introduces HGStem and HGBlock into the feature extraction module of YOLOv8 and redesigns the feature fusion layer structure to improve the detection of small targets. Additionally, the VoVGSCSP module is incorporated to replace the conventional C2F module, enhancing feature representation while reducing computational complexity. Experimental results demonstrate that YOLOv8-E achieves a mAP{@}50 of 0.85 and a Precision (P) of 0.90, showing significant performance improvements over existing models. These findings confirm the effectiveness of YOLOv8-E in accurately identifying the safety equipment wearing status of workers in UAV-captured images of construction sites.