BASIC RESEARCH ON CRACK DETECTION SUPPORT SYSTEM BY COMBINING IMAGE PROCESSING TO SEGMENTATION

Abstract

We have been studying a support system for human extraction of cracks in concrete structures from the viewpoint of image processing. The purpose of this support system is to prevent cracks from being overlooked. Therefore, we attempted to improve the detection rate of cracks by combining image processing with DeepLab v3+, a powerful method of semantic segmentation. Concretely, we examined the image of a box culvert slab taken by a UAV-mounted camera. As a result, we showed that annotation expansion and bright-ness adjustment were effective in the image processing.