Group Actions
Engagement
  • Author Author: m.ratcliffe
  • Date Created: Date Created
  • Views 767 views
  • Likes 3 likes
  • Comments 2 comments
Related
Recommended

Automated Green House Blog:5.2 - Vision based feeding Explination

m.ratcliffe

Today is a good day, The introduction of the 3rd Generation Pi is available! I cant wait to see what this does for the loop times of my vision based work on the pi.

 

See the main page for other information: Automated Green House Blog's Home Page

 

Intro

Its a great reason to continue my work on the fish feeder. In summary the feeder is a relatively advanced demand based fish feeder prototype. [We will be refining this over the next few blogs]

The Benefits of Demand based fish feeding are:

>Reduced Waste

>Increased Food Conversion

>Reduced manual labour

 

image

 

 

The ability to perform demand based feeding hinges mainly on the ability to measure the feeding rate of fish. It is critical that we are able to measure the amount of pellets eaten by fish [in pellets/second] to watch for a decreasing rate of consumption. An Idealised representation of this can be seen Below, any food delivered after a fish has a full stomach is wasted food.

 

Hardware

Figure 6 Shows the Prototype Hardware, consisting of:

>Camera [Cheap usb webcam]

>computer [Pi +OpenCv] Thanks E14 for the Pi!

>Fish Feeder

>Simulated Pond

>GUI [Touch Screen]  Thanks E14 for the Screen!

 

image

 

 

 

How It Works:

The prototype was built to achieve the following:

>Measure number of food pellets in camera view

>Perfom actuation on this information to keep pellet number above a predefined amount

 

We need to Turn a usb camera image into machine useful information, this is done in stages...

 

Performance: It works as expected in simulated conditions and a real world experiment with a large fish tank without fish.

 

Step1

Grab an image from the USB Camera

image

[Sorry Cant Rotate the image...]

Step 2

Blur image to unify colour

Step 3

Pick highest contrast colour and turn to Gray scale  image

image

Step 4

Use a Gaussian adaptive threshold technique to turn to a black and white image

image

 

Step 5

Now we have black blobs on a white background we can use a common Blob detection algorithm to identify the number of pellets in camera view.

 

 

Thats it , now we can measure the amount of food in the tank and teck how much of it a fish is eating in pellets/second...

 

Lots more work to come in this areas, Also the designs for the Pi &" case will be on-line [laser cut].

  • in reply to DAB

    hey DAB,

     

    The method you are talking about has had quite a lot of research, it works well but I would like to see a side by side comparison with inferred feeding rate [from fish activity] to actual feed rate.

     

    At the moment I turn the tank lights on, 30 seconds before feeding and they do seem to be able to link the two... But only an passing observation not a true study so far.

     

    Im drafting up a PhD proposal for the study, so hopefully there will be some real traction under this project soon.

  • Nice update.

     

    From your graph, I would think that you could use smaller doses of food and then monitor the number of fish still waiting.

    I have very limited experience with fish, but from my fishing days, I know that fish do not gather for food when they are full.

    If this holds true in your setup, then you may be able to use an extended feeding algorithm based upon the number of fish gathered.

     

    I have seen videos where people have trained their fish to respond to light or sound so that they gather for feeding.  You might be able to do a simple Pavloff study to verify if they could.

     

    DAB