Real-Time Progress Monitoring of In-Door Construction Using Deep Learning and Web-Based Application

Abstract

Successful decision-making during progress control on a construction site relies on project managers having access to reliable, timely, and intuitive information. Construction project managers and superintendents still rely heavily on manual processes, despite the fact that project control is one of the most variable and time-consuming duties. Smartphone cameras and other remote monitoring technologies can aid construction managers and other users in monitoring and controlling duties like tracking and updating project timelines. A novel deep learning progress monitoring model was built in this research employing object detection in computer vision (Faster R-CNN). The model is trained with 1000 iterations and a 0.0001 learning rate using transfer learning with ResNet101 as the base and Feature Pyramid Network (FPN) for feature extraction. By minimising the training and validation loss on the hidden test images, the models are put through their paces. Accuracy metrics including precision, recall, and overall accuracy are used to assess the models' efficiency. Based on the results, it was determined that deep learning object detection models are superior to both traditional machine learning models and deep semantic segmentation models. This thesis's overarching goal is to equip researchers and engineering experts with a workable and comprehensive Deep Learning-based system for tracking the development and deployment of indoor construction sites. In the disciplines of computer vision and machine learning, deep convolution neural networks (DCNNs) have recently emerged as a promising, stable technique. Because of its superior performance compared to more conventional image processing methods, DCNNs have become increasingly popular. Due to the rapid development of visual sensing technology (such as digital cameras), high-quality cameras are now more accessible and affordable than ever before; this has pushed research on computer vision approaches for progress tracking.viii A evaluation of the findings verifies the solutions' outstanding real-time performance with an accuracy rate consistently exceeding 75%. This research verifies the applicability of Deep Learning-based object detection and instance segmentation algorithms in construction contexts. In addition, the knowledge offered in this study can be utilised for other objectives, including productivity evaluations and managerial decisions. On a real indoor construction site, the proposed method was examined. The experimental results indicate that the approach can do automatic quantity computations in real time.