Thistlegorm: 3D Photogrammetry Survey

The technique of photogrammetry involves taking hundreds or thousands (or in the case of the SS Thislegorm, tens of thousands) of overlapping 2D photographs.

Overlap is crucial for success as the process of turning 2D images into a 3D model requires common points between each frame to tie them together. The software used for this process and to build the 3D model was Agisoft Photoscan Pro.

But its not just image overlap that is required. Photogrammetry requires every image to be in focus, and exposed correctly if you want optimum results. It is not just a question of snapping away and hoping the model will build later. Care is needed, both with the camera settings and how the subject to be scanned is approached. Any gap will mean images do not align, too many images means processing times will increase exponentially.

Data Collection

The scale and scope of the wreck was a little daunting, with the site extending over 140m long and sitting in 32m of water. The wreck can be subject to strong currents and variable underwater visibility. The explosion that sunk the ship scattered objects of interest over the seabed, including the two locomotives that reside on the port and starboard sides of the wreck. To not include these areas of interest would have left the model lacking.

By the end of the photogrammetry survey a total of 12 dives were logged and a total of 13 hours 45 minutes of in-water time recored.


Shooting enough high resolution images to cover the Thistegorm generated a lot of data. Having a strategy to organise the data was key, and every image was managed from the moment it was copied from the memory card. Processing the model began whilst on site, using Agisoft Photoscan Pro V1.3. A draft model was assembled as each dive completed. The draft model proved the images were aligning well, and would provide an invaluable reference for when the final model processing commenced.
The laptop was perfect for draft work, but the final model demanded more resource.

3D Survey Results

Final Output: Images

A total of 24,307 high resolution image files amounting to 637Gb of data was recorded during the survey.

Breakdown of how many images were recorded to build the following sections/areas of the shipwreck:

  • 15,668 images covering the seabed and external surfaces of the ship.
  • 3450 images covering deck one of holds one and two.
  • 3491 Images covering deck two holds one and two.
  • 962 images covering the forecastle rope room.
  • 1845 images covering hold three and the swim through to the break.
  • 884 images covering the captain’s cabin area directly under the bridge.
  • 532 images covering a partial model of the engine room.

Final Output: Processing

After the field work was complete post processing the data took:

  • 42 (1008hrs) days of local computing runtime.
  • 23 days (556hrs) of cloud computing runtime.
  • 65 (1560hrs) days of continuous computer processing in total.

Additional manual tasks: processing images, sorting data, reviewing model stages, uploading to cloud, downloading to cloud, backing up, organising batch jobs etc.

Internal Survey

Deck 1: Holds 1, 2 & 3

Immediately below the main deck lies the cargo holds. The first deck level runs from hold 1 just aft of the bow to hold 3 behind the bridge, a distance of 49m. Originally separated by steel bulkheads the holds are now accessible on a single level swim through.

The complex tightly packed cargo of lorries, motorcycles and trucks and low celling presented a challenge for photogrammetry. Ensuring the camera and images covered the entire space took 5165 individual photographs and several dives of careful finning.

The survey covered all three holds with the exception of the port side of hold 1. Here the main deck has collapsed onto the cargo and whilst it is possible to gain entry, the space is very tight and working with a housed camera in such an enclosed space risked damage. For that reason this area was not surveyed.

Internal Survey

Deck 2: Holds 1 & 2

Like the deck above, deck 2 presents a complex challenge for underwater photogrammetry. In both holds collapsed debris from the deck above have restricted access. The forward area of hold 1 has piles of what appear to be vehicle gearboxes, crates of Lee Enfield rifles and spare tyres,  so is quite open. However, as you move towards the stern the lorries and cargo of aircraft starter trollies and motorcycles are all packed in tightly.

Originally two separate holds, a gap has been created through the dividing bulkhead. Hold 1 at this level holds eight vehicles as loaded, with another two vehicles joining them from the collapse of the deck above.

Hold 2 has 24 lorries of various makes, and 27 Norton motorcycles, with up to a further 6 Nortons buried under the debris from hold 3. In addition to the vehicles, the hold is filled with aircraft spares, aircraft starter trollies and Lee Enfield rifles.

Scanning both holds took several dives and 3491 images. Here the depth of the wreck is between 23m and 25m, so gas consumption and decompression requirements do require monitoring.

Bridge Room


Right near the top and just under the bridge sits a small room, which is quite empty with the exception of a small bath room.

On the starboard side of the room there are a row of wash basins and central heating radiators. The main attraction in this room however is the small bathroom, complete with sink and bath tub.

Like the Rope Room near the bow, this room did not take too long to scan, needing just 884 images with most care being taken around the bathroom. It would have been disappointing to have missed the bath tub from the survey.

Rope Room Survey

Known as the Rope Room, this section of the wreck sits just under the forecastle and is accessed from the main deck. On the deck above sit the anchor winches, and the hawser pipes can be seen here.

The room was a straightforward scan for underwater photogrammetry. There are rusting drums and coils of wire here, but otherwise the room is fairly empty.

The exceptions to the simple scan approach were the cramped space in the centre of the room, and the smaller side rooms leading from the main room. In these areas the size of the housed camera meant it was not always possible to complete full coverage.

Just 962 images were needed to cover the 332m of surface area, and this was completed towards the end of a dive to scan the holds.