Estimation of Material Stocks Based on Building Structure Determination Using Image Recognition

Abstract

 Estimation of material stock of buildings in urban areas is important for recycling and proper disposal of resources. For the purpose of more accurate stock estimation, glasses and walls were extracted from images of building walls using a machine learning model of image recognition, and then calculated the percentage of glass in the walls based on the number of pixels in each. The classification of building structures based on the glass percentage showed that a threshold value of 67% is suitable for classifying steel-framed buildings with curtain walls. Classification result obtained from 262 buildings around Nagoya Station was used for stock estimation, and the result was compared with those calculated by the method proposed by previous study. The estimated amount of concrete decreased by 18% and the estimated amount of steel increased by 325%.