A Hybrid Self-Supervised Approach Towards Computer Vision Based Construction Progress Monitoring
- Publication Type:
- Thesis
- Issue Date:
- 2023
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Progress monitoring is one of the essential tasks while executing a construction project. Effective monitoring leads to accurate and timely analysis of the project’s progress which is required to make vital decisions for project control. Conventional progress monitoring techniques are error-prone, time-consuming, and require human effort. Therefore, this research aims at automation of the monitoring process of construction progress through computer vision to enable effective control of projects. This research is focused on achieving three key objectives. The first objective is to explore the state-of-the-art of progress monitoring in construction in the literature and in practice. The key takeaway from the first objective is establishing the need to work towards a robust, autonomous, and implementable progress monitoring technology for progress monitoring of construction projects. Computer vision is identified as an appropriate technology that fulfils all the essential requirements for monitoring projects autonomously. Therefore, the second objective aims to develop an integrated framework for Computer Vision-Based Construction Progress Monitoring (CV-CPM). The developed three-stage framework discusses in detail the various tools, technologies, and algorithms involved in the process of CV-CPM. The element identification stage is found to be one of the key stages of the framework. However, the existing supervised approaches require much effort in manually labelling the training data. Also, the training data cannot be reused in other projects because each construction project involves a unique set of elements, and as such, preventing them from being generalised. Therefore, there is the need for a hybrid approach for as-built modelling to overcome the individual shortcomings of the heuristics and learning-based approaches for element identification in point clouds. Finally, the third objective aims to develop a novel hybrid self-supervised approach for element identification for CV-CPM. In this context, the proposed hybrid network using deep learning based on a contrasting approach concatenated with a set of handcrafted features can extract specific features to differentiate between various elements on a construction site. This hybrid feature vector enables the network to segment various building elements from the construction point cloud data and classify them into six object classes, i.e., wall, beam, column, door, window, and slab. The model is trained and evaluated on the S3DIS dataset with the classes relevant to construction stages. The results are evaluated using the standard metrics for precision, recall, F1-score, and overall accuracy. Finally, the developed pipeline titled ‘ConPro-NET’ is tested on a mid-construction dataset.
Please use this identifier to cite or link to this item: