Automated Dimension Estimation of Steel Pipes Stacked at Construction Sites Using Euclidean Distances Calculated by Overlapping Segmentation

Abstract

Pipes are construction materials for water and sewage, air conditioning, firefighting, and gas facilities at construction sites. The quantification and identification of pipes stacked at construction sites are indispensable and, thus, are directly related to efficient process management. In this study, an automated CNN-based technique for estimating the diameter and thickness of the pipe in an image is proposed. The proposed method infers the thickness of the pipe through the difference by segmentation, by overlapping the inside and outside circles for a single pipe. When multiple pipes are included in the image, the inside and outside circles for the identical pipe are matched through the spatial Euclidean distance. The CNN models are trained using pipe images of various sizes to segment the pipe circles. An error of less than 7.8% for the outer diameter and 15% for the thickness is verified through execution with a series of 50 testing pipe images.

Keywords

• materials management
• image processing
• convolutional neural network
• perspective transformation
• image segmentation
• construction