Deep Learning -Enabled Visual Computing in Construction: Application and Digital Technology Integration

Prasad Perera^1*Srinath Perera¹Xiaohua Jin¹Maria Rashidi¹Samudaya Nanayakkara¹Gina Yazbek²Andrew Yazbek²

• ¹Western Sydney University, Penrith, Australia
• ²Commnia Pty Ltd, Sydney, Australia

The rapid advancement of Artificial Intelligence (AI) and the integration of digital technologies present transformative opportunities to improve productivity, safety, and efficiency in construction project management. This study is based on the Systematic Reviews and Meta-Analysis extension for Scoping Review (PRISMA-ScR), and 144 research articles were reviewed. The application of deep learning (DL)-enabled visual computing (VC) in construction is investigated, and a comprehensive analysis of the technological application and the DL models is conducted. While prior reviews surveyed computer vision in construction broadly, this study's systematic review focused exclusively on deep learningenabled VC and its integration with eight digital technologies through a comprehensive mapping of algorithm trends, application domains, and real-world integration challenges. The systematic analysis reveals five primary application domains: Object Detection (33%), Construction Safety (28%), Damage Detection (22%), Construction Quality (9%), and Productivity Analysis (8%). Additionally, the integration of DL-enabled VC with emerging digital technologies such as Automatic Construction Robotics, Unmanned Ground Vehicles, Unmanned Aerial Vehicles, LiDAR, Building Information Modelling, Blockchain, Intelligent Internet of Things, and Digital Twin in construction applications is reviewed extensively. An in-depth analysis of the DL algorithms and models deployed for applications revealed annual trends while illustrating the prominence of Convolutional Neural Networks and their derivatives, such as YOLO, R-CNN, Mask R-CNN, Faster R-CNN, SSD, U-Net, VGG, etc. Finally, the research identified gaps in existing research, proposing directions for prospective investigations of research gaps in areas such as real-world scalability, data quality, and ethical considerations, focusing on future work in explainable AI, edge computing, and privacy-preserving VC.