Intro / Background
This will be a relatively quick post since I did the dive on Dec 2nd and have since done Dive #7 but I wanted to provide a chronological update on my “battle” to build a photogrammetry model of the Yukon.
After Dive #5 (The “Butterfly Effect”), I had most of the wreck done but I knew that I had one very large hurdle left to finish the model: the starboard side hull near the stern. It is covered in seaweed and there is almost always at least some current / surge which moves it around quite a bit. This, of course, gives the photogrammetry software seizures since it expects objects to be “static” in nature.
Here is the status of the model after dive 5:
Here is just a random photo of the area with the seaweed:
The Dive Plan
My plan was to complete a few different tasks in addition to trying to figure out the seaweed situation.
- Better photos of the prop / rudder – I had planned on using the photos from my standalone model a while ago but they wouldn’t align so I just figured I would re-do that section and get better photos
- Stern Starboard Section – My plan here was to take pictures from far enough above the hull to use other points to align the photos and hopefully not “need” the areas that had seaweed (this ended up failing).
- Better “stern side” of rear structures – Many of the structures near amidships to the stern didn’t have good light and/or good focus photos and so they were a bit blurry. Below is an example:
Visibility was definitely not at the best. It wasn’t “bad” per se, but there was a LOT of particulate matter in the water. This wouldn’t normally be as big of a problem since the software somewhat ignores it; however, my plan of taking photos from far above the structure was completely a problem. The only way to focus was to get close.
Sample photo from dive 6:
I had a decent amount of bottom time (an hour or so) and no deco since I was close to the top of the structure for most of the dive.
Photogrammetry Model after Dive 6
Once again, I had problems with the seaweed when it came time to align the new photos. I didn’t even both making a textured model since I knew that I still had a ways to go for filling in the gaps. On that front, I had made some progress, but not a lot.
Here is the current solid mesh model after Dive 6:
Along the way, I had numerous SNAFUs while aligning photos. Here are some examples (note that these might be useful to some people who are using Metashape and trying to troubleshoot issues):
Photos 00238 – 00237
In this example, you can see how the photos are progressing along nicely from the stern in the direction of the bow and then, all of sudden, they start to spiral around each other and “get lost.”
In this example, you can see a similar problem where the photos start stacking up on top of each other.
Photos 9457 – 9515
Once again, you can see the “Spiral of Death”
Photos 9860 – 9879
And this set of photos spirals off into space:
In the model, if you look closely, you can also see part of the hull “dips down” into a depression because some of the photos are slightly out of alignment.
One of the other “tricks” I have to judge how well the photos should align into the current model (assuming I have a continuous pass of photos on any given dive and don’t have big disjoints where I skip around), I build a model with ONLY photos from a specific dive. I tried that for just the photos of dive 6 and the results were pretty hilarious:
I tried it with Medium accuracy just to see what would happen:
I’m not even sure what that is. At least on the low alignment I can make out the stern section.
In the end, I got about 65% of the photos to align correctly and filled in a few gaps and achieved the goal of “fixing” or “enhancing” the prop/rudder area but still have the seaweed problem…
5 thoughts on “Seaweed Snafu : HMCS Yukon Model – Dive 6”
I keep hearing about artificial intelligence applications creating art, or enhancing old photos and movies. Maybe this could be an application for an AI “minor dynamics omission algorithm” to look for larger consistencies by filtering out smaller dynamics in images?
Possibly. The problem is that whole sections of the hull are literally covered in moving seaweed. In some frames, literally 85+% of the photo is moving seaweed. I have tried the approach of moving further away to get other areas of the ship that are “static” but that also has problems. Underwater photography is a massive set of tradeoffs. If I get further away, my artificial light doesn’t reach the surfaces and I have trouble focussing. Or, if I boost my ISO or lower my aperture to allow more light, I get grainy photos (not a huge problem in photogrammetry but I’m already usually at ISO 2000) or I get blurry corners (the dynamics of underwater photography with domes, etc). I think the only scalable / real solution is a bit of a haircut.
That makes perfect sense. Too bad the photogrammetry software can’t work with video LOL.
It can, but there are two issues:
1) I’ve only tried it once or twice, but the software pulls frames out of the video. What I have found is that the frames it pulls are not always 100% in focus and the model tends to be a bit blurry compared to “real” photos. I have a comparison of the same target in this post:
2) The problem would still likely be the moving seaweed. The photogrammetry software expects things to be static. It might be able to account for the movement, but I think it would have a hard time figuring out which seaweed leaf was which. It will be more obvious once I post my attempt on Dive #8 in a couple weeks.