ZEMCH 2019 International Conference Proceedings April.2020 | Page 368
for computer vision‐based machine learning is suggested. A method of extracting location information
from metadata included in a digital photograph and matching it to an object in BIM model is also
clarified. Finally, the authors discuss how the extracted information can be utilized after inputting as
an attribute value in BIM data.
2.1. Construction Activity Recognitino Using CNN Models
2.1.1. Data Collection
With a recent development of computer technologies, machine learning using computer vision has
been applied to construction industries. As well as the usage of photographs, recognizing the
movement of construction equipment or worker and utilizing data for work‐efficiency and safety
management have been becoming nonspecial [12‐14]. In order to apply computer vision‐based machine
learning technology, it is inevitable to collect the appropriate amount of data and train a model.
Figure 1. The framework of project progress monitoring using digital images and BIM data: (a) CNN model
development; (b) extraction object’s information of location, WBS, and progress status; (c) updated 4D BIM data
With an aim of conducting the initial experiment efficiently, the subjects of the experiment in this
research are confined as vertical structural building elements ‐concrete wall, masonry wall, wall tile
and drywall‐. The main two sources of image collection were actual construction project documents
and the opened source of images and video clips. These images were subjected to a data pre‐processing
step of classifying and labeling as two level of Work Breakdown Structure (WBS) and their work status.
The total number of collected data as per each WBS is shown in Table 1. Experimented sources in the
dataset were originally classified as a vertical building structure of wall, and then sorted as the first
WBS of structural wall and architectural wall. And then categorized again along their materials, which
are concrete wall, masonry wall, tile wall, and drywall.
357
ZEMCH 2019 International Conference l Seoul, Korea