Monitoring and designing built environments with computer vision

The digitalization of the construction industry has led to workflows of modern construction projects relying on visual data. In particular, cameras are used for monitoring progress and construction resource activities, and Building Information Modelling (BIM) models are used to digitally represent building assets and document construction progress. The vast amounts of resulting images and videos, as well as the size, intricacy and complexity of BIM models, incentivize the use of computer vision to facilitate and automate said workflows. In addition, the uniqueness of construction site imagery and BIM structure present one-of-a-kind opportunities for computer vision research. In this thesis, we firstly explore the effectiveness of using established vision methods for action recognition for temporally categorizing and segmenting construction worker and excavator activities, and introduce benchmark datasets to incentivize further research in this direction. Secondly, motivated by the need for semantic understanding of scenes for progress monitoring, we explore means of encouraging the geometric regularity we expect to see in scenes of built environments in outputs of 2D semantic segmentation methods. Finally, we address practical concerns of existing generative models for hierarchically structured 3D shapes.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Roberts, Dominic

Contributors dc:contributor

• Golparvar-Fard, Mani
• Forsyth, David
• Hoiem, Derek
• Savarese, Silvio

• Computer Vision
• Construction Management
• Deep Learning

Statement dc:rights

• Copyright 2021 Dominic Roberts

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/113302

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/113302