3D Photogrammetry for Underwater Structures
By Cheryl Ansel, DOT bridge program engineer, NHI-qualified team leader. Reviewed by Amanda McCallister, editor.
Underwater Inspection | Commercial Diving Contractors Alabama It was late October, water temperature 8 degrees C, and the Ohio River gave me about a foot of visibility. I was trying to photograph a crack on Pier 7 of the Cutler Bridge with a GoPro strapped to a pole, hoping the photogrammetry would give the district engineer an answer without a full dive. That shadow on the sonar was just a mud seam—but I had to prove it. Since then, I've signed off on around 400 underwater inspections for state DOTs, and I've used photogrammetry on maybe a hundred substructures. It gives you a measured 3D model you can bring back to the office, zoom into, and compare year over year. Not magic, and not easy, but it beats scribbling notes on a slate. Here's how I do it, and what I wish someone had told me before my first muddy shoot.
What Underwater Photogrammetry Actually Is
Photogrammetry takes overlapping photos and triangulates common points to build a 3D point cloud. Underwater, you do the same thing, but you're fighting light scatter, silt, and a lens that behaves differently. I use Agisoft Metashape; colleagues swear by Pix4D. The software figures out where each photo was taken and builds the structure from parallax. The key is overlap. You need 80% forward overlap and 80% sidelap between consecutive shots. In clear water you can shoot from 1 to 2 m away; in the Ohio, you're lucky to get 0.5 m. That distance changes your whole dive plan. Give or take, a single pier face takes 300 to 1,000 photos just to get a model you can measure.
Field Setup: Cameras, Lights, and Scale
Pick a camera you can lock down. I use a Sony A7R III in a Nauticam housing, but a GoPro Hero 10 in a $50 waterproof housing will work if you calibrate it first. The GoPro's wide lens has heavy distortion—you'll need to shoot a calibration target like a coded sheet before you go underwater so the software can fix that. Lighting is even more critical. Water sucks out red, so without artificial light you're shooting blue-green silhouettes. I run two Sola 2500 lights angled toward the target. And never forget a scale bar: I take a 1 m flat aluminum strip, paint it with alternating black and white bands, and zip-tie it to the structure. Without scale, the model looks gorgeous but has no absolute measurements; you can measure relative distances but can't tell if a crack is 2 mm or 2 cm.
The Capture Protocol
Plan a grid for each face of the substructure. Start at the mudline and work upward, keeping the camera level and a constant standoff. You need both a horizontal and a vertical pass to give the software enough angles. The dive operator will think you're nuts, but that extra ten minutes pays off in processing. Here's the digression: if you're working in a tidal river, you have to shoot on the slack tide. Otherwise you're trying to photograph a moving train—every shift in the current ruins your alignment. I check NOAA tide tables and work only within a two-hour window.
Processing and What You Get Out of It
Back in the office, I import the frames into Metashape, align them, and build a dense cloud. Cleaning it takes forever, but the result is a point set I can bring into MicroStation or CloudCompare. I align it to previous scans and measure directly: that crack was 2 mm in 2022, now it's 5 mm. That's the kind of number the feds want to see. But don't kid yourself—photogrammetry won't replace a diver's hands. You still need to hammer concrete and feel for delamination. What it gives you is a permanent, measurable record you can revisit without getting wet. And for training, it's unbeatable: you can walk an intern around the dry side of a pier, then show them the underwater model on a screen.
Frequently Asked Questions
What accuracy can I get with underwater photogrammetry? With good visibility and a calibrated camera, I'll measure cracks to about 2-5 mm, give or take. That's fine for crack growth and scour monitoring. But if visibility is under 0.5 m, the model tends to smear. How many photos do I need? Depends on the structure and standoff. For a simple pier face, 150-300 images. For a whole pier with all faces, expect 500-1,200. I've shot 2,000 on a large truss pier. What software are you using? I'm on Agisoft Metashape. Pix4D and 3DF Zephyr are also solid. Just be ready for long processing—these models don't build themselves. Can I use a drone in a waterproof housing? A GoPro Hero 9 or 10 in a dive housing works, but watch for rolling shutter. Set exposure manually and take plenty of side-to-side frames. The wide lens needs a lot of overlap.