In this study, we aim to develop a system for detecting corrosion on the surfaces of steel bridges with high accuracy and speed by utilizing visual inspection results, and an automatic output system for visualizing the progression of corrosion. The system we aim to build consists of a multi-stage process that includes angle adjustment through image processing, detection using YOLO for object detection, and evaluation of progression using additive color mixing and image processing.
The proposed system can be divided into five stages. In the first stage, phase-only correlation is used to adjust the angle of images taken in the previous year and the current year. In the second stage, corrosion detection is performed using YOLO, a general object detection algorithm. The data applied in this study handles a large set of images, amounting to several thousand. Therefore, utilizing YOLO, which enables high-speed detection, is appropriate for processing large image datasets. In the third stage, we attempt to color the progression areas of corrosion by comparing images from the previous year using the additive color mixing method. In the fourth stage, histogram equalization and HSV space conversion are performed to evaluate the amount of corrosion progression at the pixel level.
The results of this study demonstrated that it is possible to detect corrosion and visualize the progression areas from images taken by inspection vehicle cameras using object detection YOLO and the additive color mixing method. Additionally, it was found that the progression areas colored by the additive color mixing method can be converted into the HSV space, enabling quantitative evaluation of corrosion progression.