Fish counting application

Hello, I client asked me if I could create a fish couting application. They installed a natural fish elevator and would like to measure the efficiency of it. It would be also great if we could figure out the type of fish.

Does anyone have experience with that?
Is a full HD thermal camera enough? or 4k?
Is this actually possible?

Thanks looking forward getting some inputs

Hello Pixelcrash !

Your project sound funny.

I had a similar job for a sport retail shop that wanted to count the number of visitors, and type them (male/female/young/old). This work was doing with OpenFrameworks and a simple webcam. But it was a one week project, and need a powerful computer.

So different points for your project :

  • Does your system being really precise or your must have an idea of type of fish ?
  • Is there a place where all fish must pass ?
  • Your hardware must be standalone all year long ?

In my opinion, the most difficult part is to create a hardware robust solution. Maybe without computer (but a microcontroller)

HD thermal cam can be a good input. Infrared Led too. Good Luck !

Hi Taktik,

it is going to be a permanent installation, maybe in Winter we can dismount it, I have to ask that actually.

We should get the following data:

  • Amount
  • Size (Length)
  • Type of fish

The Size can be between 4 cm up to 90 cm.
There is a ramp which is like 2 meters long and 1 meter wide, where all
the fishes have to pass.

I would like to make a concept proposal and ask for state funding to produce a working prototype and afterwards a ready-to-sell product.

I will keep you updated!

They are pretty much doing what I want

there are several already implemented method for that, mostly for tracking/counting humans, but it should work for fishes.
this should work quite well out of the box

also check the following. theres a section which describes the human tracking method used.

A thermal camera or 4K are completely overkill. I think you will get a more suitable image using IR or even RBG rather than thermal. Computer vision usualy benefits from using lower resolutions, no only because of less data to analyze but because lower res kind of smooths out the image.
There are a lot of industrial cameras for computer vision stuff. very rugged and all proof. Check pointgray cameras

As for identifying the different fish you should implement some kind of computer learning stuff. OpenCV will cover most of your needs. Artificial Neural networks should work well. but you’ll have to train it though, which can be a bit time consuming.

That’s not completely true. A while i’ve worked on a project together with an local university that could detect fish species instead of only counting. It all depends on how large the search space is and how clean the water is.

For that project, the computer had to replace a person, who counted the fish manually.