Helikite aerial sensing system: An innovative approach for monitoring construction sites.

Abstract

This study proposes an innovative Helikite-based aerial sensing system as a reliable aerial monitoring solution for construction sites. Construction site monitoring requires real-time, unobstructed, and long-duration coverage to ensure worker safety and project efficiency, but this remains challenging due to the limitations of existing methods. Key factors contributing to this challenge include: 1) fixed cameras are prone to occlusion and have a restricted range in dynamic construction environments; 2) UAVs are limited by short battery life, limited wind tolerance, and high maintenance costs; and 3) there is a lack of bird's-eye-view datasets. This study hypothesizes that the Helikite aerial sensing system can operate stably across the wind speed range typical of construction site operations. The methodology involves three core steps: 1) designing a wind-resistant Helikite-based system capable of long-duration operation; 2) constructing a specialized dataset-comprising 1,000 original bird's-eye-view images of construction sites expanded to 3,000 through data augmentation-to address the data gap; and 3) validating the system using the YOLO-v5 model for worker recognition on an independent test set of 100 images. Experimental results show the YOLO-v5 model achieved 0.9827 accuracy, 0.9850 precision, 0.9920 recall, and 0.9820 F1-score on the validation set, confirming the high quality of Helikite-captured data and the system's effectiveness.