Innovative computer vision-based full-scale timber element cracks detection, stitching, and quantification

Abstract

Timber is susceptible to environmental humidity variations, inevitably resulting in cracks parallel to the wood grain during the service life. Cracks significantly degrade the effective cross-sectional area and seriously affect structural safety and durability. Therefore, it is significant to identify the timber elements’ cracking conditions for providing reliable maintenance. Existing timber structure crack inspection mainly relies on manual work. However, with the rapid development of high-rise and large-span glued timber structure, manual-based crack inspection is not applicable to such structures for increasing workload and uncontactable high-altitude timber elements. In order to make up for the deficiencies of the existing crack detection algorithms, this paper proposed an innovative computer vision-based method inspecting full-scale timber column cracks. In step one, the crack images were stitched to exhibit the full-scale cracking condition. In step two, the YOLOv5 model was trained utilizing 425 images collected from cracked timber structures and performed K-fold crossover validation algorithm. In step three, cracking regions are quantified at the physical level. Field tests showed that the proposed method has a crack identification precision better than 0.2 mm and error below 5% compared with manual measurement, which can provide high-precision, time-saving, and noncontact in-situ crack inspection for timber structures.