Segmentation Approach to Detection of Discrepancy between As-Built and As-Planned Structure Images on a Construction Site

Abstract

Object detection has been widely used to extract visual information from resources and build components, enabling timely identification and evaluation of construction performance. However, it is difficult to evaluate the progress of a structure solely based on the detection approach. A segmentation approach using 2D images is proposed for finding discrepancies between an as-built and an as-planned structure. First, preprocessing is implemented to convert the drawing to a binary image, and a deep-learning based segmentation is conducted to extract structural components from as-built images. Then, the extracted structure and the converted drawing are compared using a 2D matrix of those images. An experiment using wood structure images was performed. The results demonstrate the accurate detection of discrepancies. Thus, the proposed approach can potentially be utilized for monitoring progress as well as inspection and maintenance, for which the identified discrepancies between as-planned and as-built images can be used.