Subsurface utility detection and augmented reality visualization using GPR and deep learning

Abstract

Recent urban revitalisation requires advanced utility management and innovative technology to achieve high-precision utility management. This paper introduces an automated framework that surpasses traditional methods of subsurface utility detection by integrating Ground Penetrating Radar (GPR), deep learning, and Augmented Reality (AR) to provide an advanced solution for subsurface detection and visualization. GPR data is collected using a multisensory GPR device, which employs antennas operating at different frequency ranges to achieve high-resolution imaging and deep penetration. Subsequently, a Mask R-CNN deep learning model is trained using a custom dataset, integrating transfer learning and data augmentation to improve detection reliability. The results are refined through profile alignment and Non-Maximum Suppression to increase accuracy. Finally, the detected utilities are visualized through a developed AR application incorporating spatial mapping and anchoring for precise model alignment and tracking. The developed system demonstrates promising results, providing an efficient utility detection and visualization solution.