Last week we set out to scan the wreck of the SMS Karlsruhe, one of the ships scuttled on the 21st of June 1919 in Scapa Flow. Everything was set; the Heart of Gold was prepared, the Suex scooter ready and the mindset was prepared for some serious scanning. The base for the week was to be the Valkryie, the Stromness-based dive boat. What could possibly go wrong?
After the first dive the images were processed and Metashape Pro started to churn the data and align the images.
Of the 3000 images shot around 80 aligned. Now normally from a batch of 3000 images we might expect 80 not to align – the duds that lack quality, overlap etc and end up being rejected. When shooting thousands of images on a single dive this level of rejects is par for the course. But the inverse – where 2920 images failed to align was unexpected.
As we have discussed before, quality of the source images is the first step for successful photogrammetry. The images looked OK so alignment settings were tweaked and run again. No change to the result. On the second dive we tried changing tactics, lighting, settings and methods but the results from the second batch of images were pretty much the same as the first.
There were now 6000 or so images refusing to align.
A Deeper Look
It was time to stop creating images and take a much deeper look into the problem. The obvious place to start was looking at the first set of images that aligned, then looking at the next-in-sequence images that failed to align.
Metashape has a very handy tool for examining the matching points. Select an image from the Photos pane and choose View Matches… from the mouse menu.
The images from the Karlsruhe were not finding matches between images that plenty of overlap. What was going wrong?
Large sections of the wreck were smothered in Brittlestarfish – like this one (the exact species remains unknown). As soon as the image of the wreck became dominated by Brittlestars, alignment stopped.
There were plenty of points being detected in the image during alignment, but few – if any – matched in overlapping frames. Why would this be?
Texture and Movement
Two additional requirements for successful photogrammetry are rich texture and a lack of movement between each frame. Photogrammetry above the waves will see alignment issues with reflective objects, or things that move between frames.
The suspicion was the Brittlestars were not only moving between frames but thanks to their uniform nature had little texture.
Were the images failing to align by an invasion of star fish? Could the project be defeated by a mass of tiny Brittlestars, all looking the same whilst simultaneously waving their little arms at me between frames?
A Helping Hand
Some truly creative minds were aboard the Valkryie and possible solutions to the Brittlestar problem started to flow:
- Capture as many 7 armed starfish (Brittlestar predators) and release onto the wreck, thus eating or scaring away the ‘problem’ starfish.
- Build a coffer dam around the wreck, drain the water and burn the Brittlestars off with flamethrowers.
- Buy a leaking Russian nuclear sub, sink it next to the Karlsruhe and let the pollution do the work.
- Turn the scooter around and use the prop wash to blow the Brittlestars off the wreck.
- Change the PH of the seawater and get the Brittlestars to move elsewhere.
- Get an army of divers to sweep the wreck clear before scanning.
With solutions like these, who needs problems? Thankfully none of the above was available in less than a week…and we know from the Cane Toad experience in Australia biological control can have unintended consequences.
Phone a Friend
It was time to pass a small subset of the images across to tech support at Agisoft. The answer came back and confirmed the suspicion; lack of texture and movement of the subject was causing alignment issues. As I write this, Agisoft are looking into a solution…
Revert to Type
For the next few days the Karlsruhe project was put to one side. Some sonar targets were examined and other wrecks scanned (more to follow in time) but on the last day it was time to look at the Karlsruhe.
A slightly modified approach yielded 1919 images of the wreck. All 1919 aligned at first pass and that number is significant, being the year the ship was scuttled…a weird coincidence if there ever was one.
The good news is large sections of the wreck images will align. Any part of the wreck not covered in Brittlestars should work just fine. This has already proved in part as the salvage break section has aligned to the section above. The better news is the salvage break section has aligned into the stern section scanned in October 2018. It will take a while to work through the 13,000 Karlsruhe images and figure out what has worked and what won’t, but we are heading in the right direction.