Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Real-Time Detection of Construction Objects on Oversized Excavation Sites by Enhanced YOLO_v7 Network Using UAV-Captured Videos
The expanding utilization of underground space has led to numerous large-scale excavation sites that conceal widespread potential spatiotemporal risk factors. Given traditional manual inspections’ time-consuming, laborious, and inefficient nature, automated real-time inspection of oversized excavation sites without blind spots is crucial for hazard identification and risk management. This paper proposes a deep learning-based model for the real-time detection of construction objects on oversized excavation sites using UAV-captured videos. Based on an oversized excavation project in Ningbo, Zhejiang Province, China, a high-quality dataset has been created using UAVs to collect multi-directional and multiscale site images under various working conditions. The NWD-IoU weighted loss function was incorporated into the state-of-the-art You Only Look Once version 7 (YOLO_v7) network to make the algorithm more suitable for construction sites. The detection performance showed a bimodal distribution, gradually improving as the IoU ratio (IR) increased. The best results were achieved at an IR value of 0.8, resulting in a 17.2% improvement in mAP@.5 and a 5% improvement in mAP@.5:95. The monitoring speed for a video with a resolution of 1920*1080 pixels is 29.97 frames per second, allowing for a precise examination of fast-moving risk actions. This study also addresses the challenges and recommends of utilizing UAVs in object detection within large excavation sites. It provides a reliable, efficient, and economic vision-based measure for real-time information collection and risk management in large-scale excavation projects.
Real-Time Detection of Construction Objects on Oversized Excavation Sites by Enhanced YOLO_v7 Network Using UAV-Captured Videos
The expanding utilization of underground space has led to numerous large-scale excavation sites that conceal widespread potential spatiotemporal risk factors. Given traditional manual inspections’ time-consuming, laborious, and inefficient nature, automated real-time inspection of oversized excavation sites without blind spots is crucial for hazard identification and risk management. This paper proposes a deep learning-based model for the real-time detection of construction objects on oversized excavation sites using UAV-captured videos. Based on an oversized excavation project in Ningbo, Zhejiang Province, China, a high-quality dataset has been created using UAVs to collect multi-directional and multiscale site images under various working conditions. The NWD-IoU weighted loss function was incorporated into the state-of-the-art You Only Look Once version 7 (YOLO_v7) network to make the algorithm more suitable for construction sites. The detection performance showed a bimodal distribution, gradually improving as the IoU ratio (IR) increased. The best results were achieved at an IR value of 0.8, resulting in a 17.2% improvement in mAP@.5 and a 5% improvement in mAP@.5:95. The monitoring speed for a video with a resolution of 1920*1080 pixels is 29.97 frames per second, allowing for a precise examination of fast-moving risk actions. This study also addresses the challenges and recommends of utilizing UAVs in object detection within large excavation sites. It provides a reliable, efficient, and economic vision-based measure for real-time information collection and risk management in large-scale excavation projects.
Real-Time Detection of Construction Objects on Oversized Excavation Sites by Enhanced YOLO_v7 Network Using UAV-Captured Videos
Lecture Notes in Civil Engineering
Wu, Wei (Herausgeber:in) / Leung, Chun Fai (Herausgeber:in) / Zhou, Yingxin (Herausgeber:in) / Li, Xiaozhao (Herausgeber:in) / Zhao, Shuai (Autor:in) / Liao, Shao-Ming (Autor:in) / Yang, Yi-Feng (Autor:in) / Wang, Wei (Autor:in)
Conference of the Associated research Centers for the Urban Underground Space ; 2023 ; Boulevard, Singapore
10.07.2024
7 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Semantic Segmentation-based Visual Detection of Construction Objects on Oversized Excavation Sites
| Springer Verlag | 2024
Layered and segmented earth excavation method for oversized foundation pit
| Europäisches Patentamt | 2023
Numerical analysis of a deep and oversized group excavation: A case study
| DOAJ | 2025