2D IMAGE SEGMENTATION AND 3D VOLUMETRIC ANALYSIS FOR AUTOMATED PROGRESS TRACKING OF ROAD MAINTENANCE AND CONSTRUCTION

MD JEWEL RANA

Road infrastructure construction projects play a vital role in economic development, regional connectivity, and social mobility. However, those projects face several challenges, such as schedule delays, cost overrun, and loss of productivity due to delayed decisions. An automated construction progress monitoring (ACPM) framework facilitates a project manager to assess the site conditions and taking decisions quickly. To develop an automated progress monitoring pipeline, this study has completed two major tasks. In task 1, we conducted a comprehensive mix-method literature review to investigate the currently available modern technologies for site progress monitoring. In task 2, this study proposed a framework to predict the qualitative and quantitative progress of a typical road construction by integrating reality capture technologies, computer vision and Machine Learning. Task 1 started with the investigation on limitations and challenges of the existing ACPM frameworks that hinder industry-scale deployment and real-world use. It aims to critically examine existing frameworks and methods, identifying their limitations and challenges, while proposing future research directions. A mixed-method review approach was employed, comprising a systematic literature review and bibliometric analysis, synthesizing data from 117 scholarly articles. This study also assessed the Technology Readiness Level (TRL) of each selected framework to evaluate its potential capability for real-world application in the construction industry. The highly segmented steps, noisy data, task-specific systems, binary progress reporting, site accessibility issues, and the lack of automated decision-making processes are emerging constraints of those existing frameworks that impede the scalable, end-to-end deployment in real projects. Moreover, most of the existing studies are performed to develop ACPM frameworks for 3D building construction which have vertical increment. In Task 2, this study has proposed a framework to automate the progress monitoring of a typical road construction. It has developed a rule-based and logistic regression-based model to predict the qualitative progress of a typical road construction project by using the extracted high-level features from input site images in two trained Mask RCNN models. It has also proposed a pipeline to quantify the amount of completed work by segmenting road layers from the GeoTiTT image of a road construction site using DeepLab V3+ based ML model. The whole methodological description of those processes is briefly described in Chapter 2. Finally, we validated our proposed framework by deploying it in a real-life road construction project in Washington State, USA. This framework has shown promising accuracies (93.73% validation accuracy and around 81.96% foreground mIoU) to be considered as a deployable framework across state Department of Transportation (DOTs).

2D IMAGE SEGMENTATION AND 3D VOLUMETRIC ANALYSIS FOR AUTOMATED PROGRESS TRACKING OF ROAD MAINTENANCE AND CONSTRUCTION

Files and links (1)

Abstract

Metrics

Details