Automated crack detection with Multi-view Stereo using high-resolution aerial images

Tomoaki EGUCHI, Nobuhiro USUKI, Naoki NISHIMURA, Yoshiaki KATSUMATA, Hitoshi KATO, Takahiro ARAKI, Yusuke YAMADA and Masafumi NAKAGAWA

Abstract

Many collapses on Mt. Fuji are caused by the development of cracks. Crack investigation of high mountain areas is generally conducted by field surveys using handheld digital cameras and tape measures. Although we use light equipments to keep safe field surveys, it is not easy to conduct crack investigations all over Mt. Fuji. Therefore, we have proposed a methodology to automated crack detection with multi-view stereo using multiple aerial images. First, we verified that our pixel-by-pixel triplet matching can achieve high success rates for point cloud generation after spike noise rejection. Second, we confirmed that our methodology can extract cracks on mountain surfaces from aerial triplet images and can draw cracks with the edge-tracking algorithm. We also confirmed that our crack drawing processing can successfully track and draw cracks, even if the cracks are unclear in images.

Key words

crack detection, triplet image matching, point cloud, crack classification