Automation of Material Takeoff using Computer Vision

Abstract

Automated material takeoff (MTO) can significantly impact construction productivity of the projects control team. The takeoff work is often a repetitive and mundane routine since it involves a manual counting of a variety of items sprawled in all kinds of locations over a drawing layout. For larger projects, such takeoffs can be time-consuming and the results can be prone to counting errors. In order to automate the task, we propose the Smart Layout Analyzer (SLA) that uses computer vision capabilities to automatically detect and recognize the items in an electrical engineering drawing layout with the aim of producing an overall item count. The software trains a Faster R-CNN with a ResNet50 convolution neural network (CNN) on the different items and their respective labels in the layout legend to subsequently localize and count the items in the drawing layout. The proposed model is different from other commercial programs that automate the takeoff making during the design process, as it can efficiently learn to count the different elements by being directly trained on the drawing layout legend.