Automatic Detection and Dimensional Measurement of Minor Concrete Cracks With Convolutional Neural Network

Abstract

Abstract. The increasing number of aging infrastructures has drawn attention among the industry as the results caused by critical infrastructure failure could be destructive. It is essential to monitor the infrastructure assets and provide timely maintenance. However, one of the crucial problems is that the budget allocated to the maintenance stage is much less than that for the designing and construction stages. The cost of labor, equipment, and vehicles are significant. Therefore, it is impossible to perform a thorough inspection by human inspectors over each asset. A more efficient method will be needed to solve this problem. This paper aims to provide an automatic approach to detecting and measuring the dimensions of minor cracks that appear on concrete structures with a noisy background. This research also investigates the relationship between image pixel size, accuracy, detection rate of cracks, and shooting distance of images. The proposed method will be able to reduce the cost and increase accuracy. A case study was performed on a concrete sewer with cracks distributed on the surface in Sydney, New South Wales, Australia.

Quantification of Structural Defects Using Pixel Level Spatial Information from Photogrammetry

“…After testing, it was found that adding Batch Norm between Convolution and ReLU could improve performance. The proposed convolutional neural network model is adjusted for concrete cracks [54].…”