Aquatic Inspection Architecture

[jhoehner]

Hello All,

I'm working on an application for a customer which uses machine vision to estimate sizes of objects in a pond. There are numerous technical challenges with this application, nearly all of which I am familiar with...

The reason for this post is to gain a little insight from those of you that have been doing this longer than I have and then relay that back to my customer to ensure we aren't making promises we can't keep. A quick summary of the application is below.

-We have a dirty outdoor pond which contains the objects

-We need to estimate length of those objects once they come into view of the camera system

-We have found that the water quality is very poor, requiring us to have no more than a 4-5" separation between the objects and the lens.

-This shortened distance requires that we use wider field of view cameras to see more of the inspection area

-We have also discussed a multi camera system with multiple wide field of view cameras. We would take the images, correct them for lens distortion, organize/stitch them into one image, and then do the image processing on that large image. 

-Lighting is a concern but the water quality is so poor that if you are further than 2-3 feet below the surface none of the suns light penetrates through, this gives us an opportunity to control the lighting we use, but the water quality does have some affect on the penetration distance of the lights we choose. 

I don't think I'm looking for answers to each of these challenges, but I am looking for some advice. What parts of this system are hard, what parts are easy, what am I missing or not expecting?

Any and all suggestions are appreciated.

Thanks all!

[cweeks12]

Here are a couple considerations I've thought of, for you to think of.

You say the objects will be within 4-5" of the camera. Is there going to be a standard distance that every object will be? Without a reference for distance from the camera, it will be very hard to know what the actual length of the object will be. You could do some sort of stereo camera setup to get that depth information if it's not there immediately.

Compounding on this, a tricky part is that most lenses you'll use have a fixed focus length. If the object that you're measuring isn't at the exact focus point of that lens, it will be  blurry and when you have a blurry image, it's difficult to measure length. 

Another consideration that you can make is to potentially use a different sort of light for your camera setup. You can look at multiple types of light and figure out which interacts least with the particulate matter in the pond. Potentially towards the IR end of the spectrum.

Overall, this looks tricky. You can probably figure it out, but it looks like it's gonna be quite a bit of work.

Also, I read your post assuming that you were looking at fish. Not sure if that's right, but that was my assumption in all this.