Investigation of Crack Detection in Mortar-Walls Utilizing EfficeintGAN

Abstract

Unlike infrastructures such as bridges, buildings become individual assets, making it difficult to obtain a dataset of damaged building images. Therefore, the application of deep learning for building damage detection using damaged image datasets has not advanced. For using the readily available images of intact conditions, this research presents crack detection methodology for mortar-finished walls of buildings based on EfficientGAN (Efficient Gan-based Anomaly Detection). In the proposed methodology, UGV, which captures many pictures or movies of indoor wall for detecting the cracks on the wall, employs the simple navigation utilizing LiDAR and AR markers. The usefulness of the proposed method is confirmed by experiments targeting university buildings.