Automatic detection and classification process for concrete defects in deteriorating buildings based on image data

Abstract

Defects in buildings accelerate the deterioration of building conditions and threaten the structural safety of buildings. Building condition assessments are essential to identify and assess building defects. However, with the vast number of buildings, individual inspections are nearly impossible, highlighting the need for innovative solutions. Therefore, this study presented a safety inspection process that automatically detects and classifies defects using image data that can be easily acquired at a low cost and confirms the location of defects through 3D modeling. A CNN model is developed to detect defects in images of aging buildings, and the proposed process is validated through a case study to demonstrate its effectiveness. The case study focused on the effectiveness and convenience improvement of the proposed process compared to the existing methods of inspecting the exterior of buildings. The effectiveness of the proposed process was reviewed through the case study, and it was confirmed that more efficient safety inspections could be supported using only image data compared to the existing methods. In particular, the safety inspection process of this study is expected to be applied in practical use within a short period since it is an easily utilizable method for inspectors.

Keywords

• concrete defects
• safety inspection
• defect images
• convolutional neural networks (cnn)
• building defects location