Journal
AUTOMATION IN CONSTRUCTION
Volume 147, Issue -, Pages -Publisher
ELSEVIER
DOI: 10.1016/j.autcon.2023.104745
Keywords
Crack detection; Tiled sidewalk; Deep learning; YOLO; Computer vision; Unmanned aerial vehicle
Ask authors/readers for more resources
This paper proposes the integration of You Only Look Once (YOLO) into an unmanned aerial vehicle (UAV) for real-time crack detection in tiled sidewalks. Different network architectures of YOLOv2-tiny, Darknet19-based YOLOv2, ResNet50-based YOLOv2, YOLOv3, and YOLOv4-tiny are compared to improve accuracy and speed of detection. The results show that ResNet50-based YOLOv2 and YOLOv4-tiny offer excellent accuracy and speed, and remarkable ability in detecting small cracks. They also demonstrate good adaptability to environmental conditions such as shadows, rain, and motion-induced blurriness. The evaluation suggests the appropriate altitude and scanning area for the YOLO-UAV-based platform to achieve remote, reliable, and rapid crack detection.
The conventional method of manually verifying the quality of tiled sidewalks is laborious, because of the time-consuming identification of cracks from numerous grid-like elements of tiles. In this paper, the integration of You Only Look Once (YOLO) into an unmanned aerial vehicle (UAV) is proposed to achieve real-time crack detection in tiled sidewalks. Different network architectures of YOLOv2-tiny, Darknet19-based YOLOv2, ResNet50-based YOLOv2, YOLOv3, and YOLOv4-tiny are reframed and compared to get better accuracy and speed of detection. The results show that ResNet50-based YOLOv2 and YOLOv4-tiny offer excellent accuracy (94.54% and 91.74%, respectively), fast speed (71.71 fps and 108.93 fps, respectively), and remarkable ability in detecting small cracks. Besides, they demonstrate excellent adaptability to environmental conditions such as shadows, rain, and motion-induced blurriness. From the assessment, the appropriate altitude and scanning area for the YOLO-UAV-based platform are suggested to achieve remote, reliable, and rapid crack detection.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available