An Image-Based Model Framework for Automatic 3D BIM Generation of Existing Buildings Using Synthetic Images

Abstract

The purpose of this study was to develop an image-based framework to generate 3D parametric BIM of existing building using Mask R-CNN and Autodesk Revit with Rhino Grasshopper. This study accomplished the following objectives: 1) development of a Mask R-CNN model to predict and extract objects’ type, location, and mask in image, 2) development of a 3D BIM generation framework using Autodesk Revit and Rhino Grasshopper to create 3D BIM based on the output of the Mask R-CNN, and 3) synthetic data performance analysis to optimize the Mask R-CNN training process, which is the main task in the image-based framework. Efforts have been made to make existing buildings into BIM for effective maintenance, but it suffers from insufficient documentation of building information. In addition, BIM creation of current buildings is mainly performed manually, which is inefficient in terms of time and cost. Although some efforts have been made for automation, most of them require excessive specialized knowledge or complex equipment and analysis. Therefore, there is a need to develop a novel framework for automated BIM creation of existing buildings that is easy to use and more effective. The results of the developed image-based framework show that 3D parametric BIM of the existing building including inferred hidden structures can be created from images by integrating the deep learning, BIM technology, and architectural domain knowledge. The image-based framework is expected to help architects and engineers to obtain insights and rationale for the successful generating 3D BIM of existing buildings, especially in applications of hardening large numbers of structures against natural disasters.