Automated Construction Progress Monitoring using Image Segmentation and Building Information Models
Open Access
- Author:
- Jiang, Zhouqian
- Graduate Program:
- Architectural Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 13, 2023
- Committee Members:
- Rob Leicht, Major Field Member
John Messner, Chair & Dissertation Advisor
Julian Wang, Professor in Charge/Director of Graduate Studies
Yuqing Hu, Major Field Member
Yanxi Liu, Outside Unit & Field Member - Keywords:
- Construction Progress Monitoring
Building Information Modeling (BIM)
Computer Vision
Image Segmentation
Deep Learning - Abstract:
- Construction progress monitoring plays an important role in understanding the current status of a construction project and ensuring the project is delivered on schedule. Traditional progress monitoring approaches relied on manual inspections and paper-based documentation, which were time-consuming and prone to human error. With the advancements in digital photography platforms and computer vision techniques, visual data documenting the actual construction progress are more accessible to project teams, and the essential information can be automatically interpreted and extracted by computer vision-based approaches from the acquired visual data. The increasing adoption of Building Information Modeling (BIM) further enhances the capabilities of identifying completed elements and detecting deviations from the as-planned schedule of various construction activities. This research aims to develop a process model for automated construction progress monitoring using image segmentation methods and BIM. To achieve this goal, a systematic literature review of journal papers was first conducted to establish the theoretical foundation of this research. The reviewed papers focus on computer vision applications in the construction and asset management phases. Then, to investigate the implementation feasibility of image segmentation methods in the construction field, a case study was conducted to implement three popular image segmentation models (U-Net, DeepLabV3+, and Mask R-CNN) to segment structural steel elements in a three-story steel structure project. Lastly, a 2D image-based process model was developed to identify the completion status of construction elements associated with specific construction activities by pixel-wise comparison. The process model can be achieved with limited images captured onsite and can accommodate different image-capturing methods and image segmentation models across various construction activities. This research categorized all the reviewed papers into five construction use cases and six computer vision domains. The trending popularity of each construction use case and computer vision domain within the reviewed period was analyzed to inform future research and development in this interdisciplinary area. Cross-validation was used to train the three image segmentation models on a self-developed open-source dataset. The performances of the three models were quantitatively evaluated, showing that DeepLabV3+ achieved the best performance. In addition, this research proposed novel workflows within the process model to address the occlusion and overlay issues caused by other construction objects. The workflows were validated on a real construction site, indicating a threefold improvement in occlusion scenarios and a 50% reduction in overlay scenarios.