Underground mines are some of the most hostile environments for spatial data collection, where GPS signals vanish and every centimetre of rock movement can signal risk. Yet the last decade has seen a quiet revolution in how miners map these spaces, with drones equipped with laser scanners flying into stopes—those vast, irregular cavities left behind after ore is extracted. These unmanned aerial vehicles (UAVs) now generate dense 3D point clouds at regular intervals, offering unprecedented visibility into ground stability. But there’s a catch: without GPS, each scan lives in its own arbitrary coordinate system. To make sense of change over time—critical for safety and planning—engineers must manually align these clouds with the mine’s global reference frame, a painstaking process that slows everything down and risks inconsistency.
Now a team led by researchers at the University of New South Wales has developed an automated georeferencing system that could eliminate much of that manual drudgery. Their method, called 3D Tag-based Automated Registration and Georeferencing Technique (3D-TARGeT), uses simple rectangular fiducial tags—cheap, generic, and non-unique—placed strategically around the stope. These tags act as fixed landmarks visible to the scanner, allowing the system to automatically match each new scan to the mine’s master coordinate system. No human alignment required. In tests using four multi-temporal point clouds from an actual underground stope, the system delivered centimetre-level accuracy, with median errors below 3 centimetres across all scans. It outperformed traditional automatic registration methods by a clear margin, proving both robust and reliable.
At its core, 3D-TARGeT combines three key steps: tag detection, geometric matching, and rigid transformation. The scanner captures the scene, identifies the tags using computer vision, extracts their positions, and computes the precise shift and rotation needed to align the new data with the mine’s reference network. Because the tags are low-cost and easily printed, they can be deployed in large numbers without financial burden. Their non-uniqueness is actually an advantage—it simplifies tag production and placement while still allowing reliable matching through geometric constraints.
For the mining industry, the implications are significant. Routine geotechnical monitoring—tracking deformation, stress redistribution, and potential collapse—relies on accurate, timely data. With 3D-TARGeT, mines could process laser scans overnight, feeding geotechnical models with fresh, georeferenced point clouds without manual intervention. This reduces downtime, increases data frequency, and improves safety by enabling real-time hazard detection. Importantly, the accuracy level achieved (centimetre-scale) is sufficient for most geotechnical analyses, including convergence monitoring and structural analysis of rock masses.
But the application extends beyond traditional mining. Defence and security sectors operating in subterranean environments—such as tunnel surveillance, counter-IED operations, or underground facility monitoring—face similar challenges in mapping GNSS-denied spaces. Autonomous drones or ground robots equipped with LiDAR could deploy 3D-TARGeT-style tagging to build and update 3D maps of tunnels, bunkers, or urban warfare zones. By automating georeferencing, these systems could support faster mission planning, persistent surveillance, and improved situational awareness in denied or contested environments.
The research team’s proof-of-concept demonstrates that automation doesn’t require expensive bespoke solutions. A framework built on generic tags and standard algorithms can deliver industrial-grade accuracy. As underground mining pushes toward fully autonomous operations and defence agencies prioritise real-time, accurate 3D mapping in complex environments, tools like 3D-TARGeT could become foundational infrastructure—turning raw point cloud data into actionable intelligence, faster and more reliably than ever before. Read the original research paper here.