We at EyeROV during our ROV underwater inspections had faced problems like colour loss, image degradation in murky waters etc… Most of this degradation is due to wavelength-dependent absorption and scattering including forward scattering and backward scattering. We have developed and implemented multiple underwater video/image enhancement algorithms to solve this problem. We have used a combination of different AI-based enhancement techniques to develop our software, which can be applied in different underwater conditions. Check out our sample video and enhanced results here
Always cool to see new advancements in this space
A few questions:
- Are the three samples using different technologies/algorithms? If so, what are the expected differences between them?
- Are these available to trial or use commercially at this point? Should interested people ask questions here, or is there some way you’d prefer for them to contact you?
- Are these currently real time, or post-processed? If realtime, is it run in the ROV or on the topside?
- What hardware are you using/is required to run the processing?
As a side note, I’ve edited your post so the videos are embedded, which means they can be watched from within the forum. There are instructions for doing that in the Formatting a Post/Comment section of the “How to Use the Forums” pinned post
Yes, Different algorithms are used . The selection of processing algorithm depending on the water visibility level.
They are post processed .
Currently we are running it on a desktop PC.
Alright cool, thanks for the responses!
Interesting, I suppose that makes sense. Do you have some automatic way of selecting that (possibly even using different algorithms within one video), or are they manually switched between at this point?
Good to know
Do you have some kind of time indication for roughly how long it takes to process a video? Of course that depends on the computer hardware used and the video resolution and duration, but it’s useful to understand if processing a 1hr video is expected to take 2 hrs vs something like 2 days
Separately, is real-time something you’re working on achieving, or is that not part of the plan? It’s understandably useful to be able to see the processed result as you inspect something, but that also generally requires specific hardware and/or a decent amount of algorithm optimisation.
You can ‘quote’ parts of a comment by selecting it with your mouse and pressing the “Quote” button that appears, as covered in Interacting with Posts
Sorry for the delayed response.
As of now we are manually doing the selection . An automatic algorithm selection procedure is under development.
It depends on the selected algorithm.our algorithms typically take 2x to 10x of the video time
Also we are trying to make it real time one.
Thank you .