How to 3D Scan an Underwater Cave

Every few weeks someone writes to me with the same question: "I dive caves. I want to scan them. Where do I start?" Well, the scanning itself is the easy part. Reconstruction and presenting it to broader audience is where it becomes more tricky.

Here is how we did it in the flooded Maria Concordia mine in Poland: what worked, what broke, and what I would tell you to do first.

Start with why, not with the camera

Photogrammetry is a tool, not a goal. Before you plan a single dive, answer two questions: why am I scanning this place and what will I find there?

• A map. If the site is a mine with long, straight corridors, a 3D scan is the wrong tool. A laser distance meter, a compass and a wet notebook give you a better map in a fraction of the time. Scanning earns its keep on irregular shapes: caves, collapses, restrictions, shafts.
• Safety. A model that divers can "swim" before the dive removes one stress factor from an unknown site. Mikko Paasi told me about a restriction in Tham Luang, during the rescue, that the team squeezed through blind. When the water dropped it turned out to be a keyhole. The line ran through the narrow part.
• Change over time. Glacial caves and wrecks change from year to year. A scan repeated on a schedule becomes a measurement, not just a picture.

If your answer is "because it's cool", that is fine too. Actually that would be my answer in most cases. Just know that it decides everything downstream: resolution, coverage, how many dives, how many batteries.

*Disclaimer, so you don't say I'm a terrible person: Be a cave diver first, a scanner second

Scanning an overhead environment takes all of your attention. Gas checks, buoyancy, line awareness and light management have to run on autopilot, because the camera rig will take the rest. My own path was about five years from recreational diver to full cave, and another three years before my first mine scanning project. I am glad I gave myself that time.

Two things will end a scanning dive early if you are not ready for them:

• Silt. In Maria Concordia the sediment is so fine that the water moved by a passing diver lifts it. Add regulator bubbles knocking particles off the ceiling, and you must assume the way back will be in zero visibility.
• Cold. The water is 8 °C. Two of our four divers found drysuit leaks on the first day. Spare suits saved the project.

If you are not cave certified, you don't have to wait until you are. Spend this time wisely, by simply practicing in dry sites first: caves, bunkers, old fortifications. A phone and open-source software are enough to learn the craft.

And when the time comes, you'll be ready for the fun part.

The rig: a boom, not a handheld camera

The first thing I learned, on a small test site called Marie Agnes (under 100 m of flooded corridor), is that holding the camera in front of you the way an underwater photographer does simply does not work. You cannot sweep a corridor cross-section from your hands.

What works:

• A 1.5 m boom with the camera and a single wide-angle video light at the end.
• Balanced to neutral with foam, so the whole set weighs nothing in the water.
• The camera always perpendicular to the surface you are photographing. Without moving your body you can trace a full circle around the corridor and cover walls, ceiling and floor.
• A bonus: the camera sits ahead of you, in clearer water, before your fins stir anything up.

For Maria Concordia we needed faster coverage, so the boom carried five GoPro cameras, angled so that each field of view overlapped the next by about 50%. Shooting one frame per second, the set captured roughly half the corridor cross-section every second. Lighting came from four wide-angle video lights.

Overlap is the whole game

Photogrammetry software builds the model by finding the same points in many photos. If a patch of wall appears in only one frame, it does not exist. The rule on every dive: every part of the wall, ceiling and floor must appear in several shots, from slightly different positions. Move slowly, keep the distance to the wall steady, and never swing the rig faster than the light can follow.

The problem is not taking photos. It is knowing what you have already covered. A a mile of corridor looks the same after the fourth pass. We kept a working map on the surface, marked each scanned section after every dive, and briefed the next diver from it. Systematics beats heroics.

The surface team runs the show

Here are the numbers from our first two and a half days in Maria Concordia. They explain why this is not a one-person project:

• 4 cave divers, 2 people on surface support, 2 photogrammetry and 3D specialists
• 5 cameras and 4 lights, with 3 full battery sets for each
• 48 light batteries and 15 camera batteries in rotation
• 6 scanning dives a day
• about 110,000 photos across 4 flooded levels

After each dive the rig went straight to the surface team: batteries swapped, cards copied to disk, a fresh set handed to the next diver already waiting in the water. One detail that matters: the "dry" chamber at the bottom of the 40 m shaft is so humid that we never opened a camera housing down there. Moisture trapped inside a GoPro case condenses in 8 °C water and fogs the lens from the inside. Every battery change happened topside.

The logistics are way more simple if instead of traditional cameras, you use a single or double 360 camera. I'll be describing this workflow in another article.

Plan B, C and D

Cave training teaches you to carry a backup for everything. Scanning projects need the same mindset. Before the first dive we had lost our underwater photographer to illness, several memory cards refused to work with the rented cameras, and one camera died without explanation. We bought cards on the way, reconfigured the rig around the cameras that worked, and rescheduled the photo session for a later trip. None of it was dramatic, because none of it was a surprise.

What happens after the dive

Processing the material from that first trip took well over a month of computer time. Then came the part nobody warns you about: a raw 3D model is not a story. Nobody outside cave diving will look at a point cloud for more than ten seconds.

What people did look at were the cross-sections: the mine sliced open, magnesite veins on the walls, the remains of scaffolding and winches still in place. Then the fly-throughs, then the VR experiments. The scan was the beginning of the project, not the end.

That is exactly the gap U3DA exists to close. You already know how to plan the dive and run the camera. The course takes the model you bring up and turns it into renders, cave fly-throughs and finished dive videos in Blender, using the workflows from these same projects. The first lesson is free.

Further reading: my full account in InDepth magazine, the Maria Concordia project at the Submerged Foundation, and the XDEEP Exploration Support Program project page and field story.