Integrating GPR Inversion and Visual SLAM for Underground Infrastructure Detection and Localization.
01 — Problem & Rationale
The Problem
Underground utility networks are often poorly documented, creating risks during excavation and infrastructure construction. Ground-penetrating radar (GPR) provides subsurface sensing capabilities, but interpreting radar signals and accurately locating buried utilities within physical environments remain challenging.
Our Rationale
We integrate learning-based GPR inversion with vision-based spatial mapping to reconstruct subsurface properties, characterize buried utilities, and register their locations within the surrounding environment.
GPR sensing principle: a moving transmitter and receiver survey a buried pipe, forming a curved reflection in the radargram.
02 — Methodology
We develop an integrated framework that connects radar data generation, subsurface reconstruction, and spatial localization.
2.1 — Learning-Based GPR Inversion
We use a generative adversarial network (GAN) to enhance simulated GPR radargrams for model training. A deep learning-based inversion model then reconstructs subsurface permittivity distributions from radar measurements, enabling the estimation of buried pipe locations, depths, and diameters.
Simulated radargram generation, GAN enhancement, and learning-based reconstruction of subsurface permittivity.
2.2 — Vision-Based Spatial Registration
We integrate GPR-derived pipe detections with camera-based simultaneous localization and mapping (SLAM). By estimating the GPR survey trajectory and reconstructing the surrounding environment, detected underground utilities can be spatially registered within the site map.
Real-world experiments further demonstrate the feasibility of integrating GPR-based subsurface detection with vision-based spatial localization.
Research Implications
This research provides an integrated approach for transforming subsurface radar measurements into spatially registered utility information, supporting more informed excavation planning and underground infrastructure management.
