In a team of Capital High Technology Co., NAIST and Tohoku University, a machine learning system for concrete crack detection was developed using image analysis of characteristic patterns specific to cracks. Cracks of 0.2 mm or more were detected with high accuracy of 80% or more even when covered by dirt or scratches on the surface. Working time was shorten from about 300 to 30 min.

NEDO news release, August 3, 2017