I thought it was a good place to ask my question since some of you are very active. I’m a researcher working on geological problems that involve underwater karst systems.
I’m seriously considering buying a Trident to perform some photogrammetry to generate 3D models of different types of cave. I know it’s possible and feasable to do it on sea or lake bottom when the surface is kinda flat but do you think it still works if the ROV has to take pictures of the cave roof ? What possible technical issues could we face ?
I am also interested to do 3D cave mapping.
As we started using OpenROV to explore deep siphons in our caves, we also need to make documentation.
I would love to be able to make a recording and build a 3D model based on one run , as diving in the siphon (based on the last year experience - manouvering, water flow, underwater rocks, etc) is not an easy task, Especially coming back, when the water is not clear anymore.
From the surveying point of view simple telemetry is not enough as the quality of data degradates quickly. My cave partner who is a professional surveyor is never happy enough with the measurements. The solution is to use different data from other sources.
ROV can make pictures (even tons of them), record the movie and also telemetry, but the issue is to make 3D based on all data, first we would need a common timeline for all of them (which is probably achievable as pics and movies have timestamps, second the data from all sources need to be processed to the 3D model. There is SLAM algorithm used by professionals to make 3D mappings:
This is what I would like to achieve with OpenROV.
It should be possible to align all data coming from pics, movie and telemetry as BBB and arduino record all events in the logs but I did not figure it out yet.
Another idea is to to use some kind of lidars attached to OpenROV , but they are expensive and put additional load to OROV. The solution can be to tow it, but the price is still hight, this solve just one issue partially.
So the idea of using what is already on board is very tempting.
Thank you very much for this complete answer Jack !
In my opinion several problems are to face here in order to do photogrammetry:
First, the quality of the acquired images. In a large cave where the exit is still visible, the progresive lost of red and yellow can be fixed by attaching lights to the ROV (still putting additional weight to it anyway). In a siphon the density of particles in a water can’t be solve I guess except if you do as you say, acquire all the data on the way to the bottom.
-Second, the GPS tracking of the ROV through the cave. Once again, in an open cave a Navimate wrist device could be interesting to use if it’s attached to the ROV (https://www.navimate.com/). In a siphon that remains a problem. There’s a tremendous variety of SLAM algorithms and they seem quiet complex for a non mathematician. Do you think a laser technology, which is rather light, could be useful ?
@lucasvimpere I’m a caver and have done 3D modeling in the sea and outside of a cave, through photogrammetry. The issue I’ve found with doing the 3D models in the caves is lighting. The moving of the light source for the images cause variations in the shadows, which will make it very difficult for the software to align the photos correctly. Though I’ve had some success in smaller caves, the larger caves have been much more difficult. Using my Trident to do the same has been an even greater challenge since the only available light source is the LEDs on the front of the Trident.
Right now my team and I are working on building a sunsphere to illuminate the areas we wish to model. We’re hoping to get it to put out 90,000 - 120,000 Lumen. We’ve had great success converting video to 3D models, so due to the limited battery time of the sunsphere, we expect that would be how best to collect the images for the models. Using the sunsphere to follow an ROV would be difficult though. The only plan we have now is to setup stations for the sunsphere to be anchored along the flooded passage, and do a series of video runs to collect all needed data. Due to battery issues, this might require many trips to the cave to do the work.
Sounds like a good solution to the problem. I’m currently working on a very large cenote so light variability is not really a problem as the turbidity of water since it’s a karst that goes through carbonate units only. The light sphere remains a solution where cave divers or cavers have access. Our goal is to use a ROV to go where we can’t because of the size of siphons, to check if the cave is connected to the sea. A very good paper on edge detection based on thresholded images of a video sequence has been published, using stereo vision though (“Underwater cave mapping using stereo vision” Weidner et al., 2017).
Another paper : “Sonar visual intertial SLAM of underwater structures” by Rahman et al., 2018 is also a promessing method that doesn’t seem too expensive. I know that it’s a homemade device so maybe @acaconca you could ask them which algorithm they use to solve your SLAM problem ?
@lucasvimpere Thanks, I’ll be reading this paper when i get home tonight! Years ago I had looked into doing 3D models & mapping using stereo images, but since I was so much success with photogrammetry, I never followed up on stereo vision.
Where I go light is always an issue. The most recent spaces I’ve sent the Trident into are so large that at times the ROV sees nothing but particulates in the water surrounded by an inky blackness. Hopefully the SunShpere will fix that issue. Problem solving is half the fun of exploring these space!
I am impressed: 120,000 lumen light source.
I my case it is unrealistic. ROV cannot be assisted by a diver, it would not make sense for us. On our stage of exploration we use our ROV to do reconaisances of siphons to verify if it is worth to put efforts and engage a divers. The reason is a long distance to the target. And the risk of course.
So that’s the reason I am looking for the solution to use just ROV.
@acaconca Well don’t get to impressed yet. Though 120,000 is possible, right now its just a dream for us. There are constraints like cost, energy consumption, portability, heat, and many more. Right now we’re shopping for the parts to assemble a single panel of the six or more we’d like to have in the sphere. The results of our first build could change the plan, and decrease the lumen output greatly.
We’re lucky and have two very experienced Navy/Cave divers for the current project. Since we’re not going through any space that should be tough for them to explore right now, we can enjoy the luxury of the sunsphere.
In in the summer of 2019 we’ll have new challenges though. We will be sending the ROV into spaces cavers haven’t been able to get past, and I think the solutions you’re examining would be very helpful. I’m very interested in seeing what you come up with and how things go for you.
@oddmarjacobin, I’m working on a light using several of those LED panels. My thought about heat was the same. My plan is to have a series of them mounted together in a semi-spherical shape. I have all the components already, but just haven’t had the time to assemble them.
My plan is to lower the light above the area of exploration like a Mini Sun. Since the lights on the ROV will change the shadows on the place being examined, I’ll need to turn off the lights on the ROV. Hopefully the Mini Sun will produce enough light to illuminate the area from a fixed location, and allow consistent images for Photogrammetry.
As the observer (ROV) moves, the apparent shadows will shift. Having multiple MiniSuns might eliminate most of the shadows.
I know from having just one light in my welding shop, it was always casting a shadow where i couldn’t see well. After putting one light in each corner, the shadows are now all but gone.
SLA AGM battery (12v 10aH from Amazon) on a three-legged stand, with a meter of PVC going up to the MiniSun lights.