01 April 2019-31 March 2020
Project completed
What was the goal? Every year lack of real-time feedback on construction progress and quality is the main cause of delays and cost overruns worth billions of dollars in the global construction industry. Through application of cutting-edge technology, this project aimed to equip the construction industry with real-time insights on construction progress, installation quality and thus help deliver large construction and infrastructure projects on-time, on-budget and with enhanced productivity. This project aimed to develop a prototype to process visual and sensed data for construction progress monitoring in a fully automated fashion, a process known as scan-vs-BIM (Building Information Modelling). Our research employed BIM and computer vision methods, in addition to exploring the potential of deep learning in this context. Automating manual progress and quality reporting processes in the construction industry allowed to identify problems early on and respond quickly, avoiding reworks and overruns of cost- and time-schedules. Without these automations, it could take up to several weeks for critical information to reach stakeholders.
What did we work on?
The main objectives of the project were:·
- A comprehensive implementation and especially deep-learning focused review of the problem of scan-vs-BIM from visual and sensed data.·
- The proposal of a new approach to progress tracking using recent methods from the fields of computer vision, robotics and deep learning.
- To achieve our objectives, we assembled a unique team of experts to conduct state of the art research on 3D computer vision, Building Information Modelling (BIM) and machine learning techniques, contributing to the advancement of cutting-edge technology and the UK’s research leadership in the built environment.
What did we do?
The project yielded several outputs:
- The report Deep Learning for Construction Progress Tracking from Visual and Sensed Data.
- Laser scanned, synthetically augmented construction site sandbox dataset.
- Scan-vs-BIM Python library (SvBL) and prototype application for semantic interpretation of construction site laser scans.
The collaboration between the Centre for Systems Engineering and Innovation and the construction firm Contilio ensured both academic rigour, cutting-edge research outcomes, and applicability. We worked together with Contilio on creating a construction site sandbox dataset from real-world and synthetic 3D point cloud data, obtained from terrestrial laser scanning. Our findings had direct dissemination and impact to industry thanks to a presentation of our work to an academic and industry audience at the 2019 CSEI - Industry Showcase event, hosted at the Institution of Civil Engineers (ICE). The great interest around our findings granted us further funding from Innovate UK as part of a direct follow-up project round.
How to engage with us
If you want further information about the project or if you want to get involved contact us csei@imperial.ac.uk
Funding information
The project was funded by Innovate UK. The research was carried out at the Centre for Systems Engineering and Innovation in collaboration with Contilio.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
01 April 2020 – 31 October 2020
Project completed
What was the goal?
This project was a direct continuation of previous research at the Centre for Systems Engineering and Innovation and aimed to develop a prototype to process visual and sensed data for construction progress monitoring in a fully automated fashion. Our research employed Building Information Modelling (BIM) and computer vision methods, in addition to exploring the potential of deep learning and semantic SLAM in this context. Automating manual progress and quality reporting processes in the construction industry allows to identify problems early on and respond quickly, avoiding reworks and overruns of cost- and time-schedules. Without these automations, it can take up to several weeks for critical information to reach stakeholders. Ensuring real-time feedback on construction progress and quality through the application of cutting-edge technology means tackling the main cause of delays and cost overruns, which cost the global construction industry billions of dollars.
What did we work on?
Our main aim was to build on the findings from our previous project. We set out to do this by carrying out:
- A review of Scan-to-BIM for progress monitoring and a comparison to Scan-vs-BIM approaches.
- The development of a prototype for large-scale reconstruction and semantic understanding from image data.
What did we do?
Our research findings on Scan-vs-BIM, Scan-to-BIM and semantic scene understanding within the scope of this project are applied to and continued within the project Design change in digital twins, as part of the data-centric engineering program of the Alan Turing Institute, a sign of the academic robustness and innovative nature of our findings.
The project produced several deliverables, including:
- The report Scan-to-BIM for automated as-built modelling.
- Extension of the Scan-vs-BIM Python library (SvBL) to read E57 laser scans.
- Development of a Convolutional Neural Network approach for semantic segmentation of construction site laser scans.
- Evaluation of our approach on the construction site sandbox dataset from the previous project.
The findings had real-life impact as we worked together with Contilio on:
- Extending the construction site dataset that originated from the first project with semantic annotations from BIM data.
- Deploying our semantic segmentation model as part of Contilio’s internal R&D on automated construction progress tracking.
How to engage with us
If you want further information about the project or if you want to get involved contact us csei@imperial.ac.uk
Funding information
The project was funded by Innovate UK. The research was carried out at the Centre for Systems Engineering and Innovation in collaboration with Contilio. Research was conducted jointly at the Smart Robotics Lab.
