toddak

So I've just invested in the cyclops objects, which looks like it can do all sorts of powerful things. Its just the how I'm not sure of yet...

As a good solid starting point, I have two things i'm trying to figure out how to utilize cyclops and a web cam for. Namely:

How can I get cyclops to tell me (or initialize a bang) when something new enters the web came view? Or specifically, I mount a web camera in a gallery space, and want sound to trigger when someone enters?

And secondly, how can I get it tp tell me if there is movement in a zone? Or specifically, omit a bang (to trigger a sound) when someone moves in the space, but not when they are standing still?

I've done a few forum searches, but can't seem to answers to these questions. I've also tried contacting Eric SInger via his website for more information, but the emails are being returned :(

the-wonderful-world-of-cyclops-need-a-little-help

I don't know anyone who (still) uses cyclops, maybe you'll like the free openCV library: 

I don't know anyone who (still) uses cyclops, maybe you'll like the free openCV library:  http://jmpelletier.com/cvjit/


toddak: Okay let's break it down into several different steps and see if Cyclops will help you in what you are trying to do.

http://www.flong.com/texts/essays/essay_cvad/

For your second part, you can use frame differencing for movement only detection.

Frame differencing can easily be done like this:

Gives you movement, but not presence *(lots of ways to do this, lots of variants on how to get a more accurate output, cv.jit.sum is tailor made for this).

The kinect gives the most stable and cost effective presence there is at the moment, but has it's limitations.

One way Cyclops can help you is if you wanted to test if something is in a zone by testing if it is NOT the color of the floor that you have pre-sampled. It'd be just like background subtraction except that instead of testing for luminosity you are looking for color. It's by no means bullet proof, if the person is wearing something too close to the color of the floor, you will get a false negative.

Cyclops is cool, it's the most powerful thing for specifically doing color tracking, but as far as just basic CV, you would be better off exploring combinations of techniques before deciding what works best for your purpose.

For starters, read this:
http://www.flong.com/texts/essays/essay_cvad/

One way Cyclops can help you is if you wanted to test if something is in a zone by testing if it is NOT the color of the floor that you have pre-sampled.  It'd be just like background subtraction except that instead of testing for luminosity you are looking for color.  It's by no means bullet proof, if the person is wearing something too close to the color of the floor, you will get a false negative.

@ Timo: Thanks for the link to the cv.jit packs. I've had a look at them before, but perhaps I was looking for an obvious object (as opposed to using my brain!). But shall revisit them now, have progressed a bit in max since then!

@cap10subtext: Wow, what an amazing article! Plan on reading that on my train journey shortly. Thank you so much for that one!

The idea of frame differencing was something I was trying to calculate to begin with. I knew that mathematically speaking, if you took a base image of an installation space from the web cam, and then compared that with images taken from the web cam at various intervals, then you should be able to compare the matrix's. But I never got around to coding it. I wouldn't say it was laziness this time, but just that I wasn't sure if I could program it accurately enough which is why I started to look into alternatives like cyclops.

The jit.freenect.grab object looks amazing, and in an email with Rafael Lozano-Hemmer (whom I notice appears in that essay also!) he was recommending the use of kinect's too. They do seem quite powerful for the little you spend on them.

Would I be right to assume by looking in at the jit.freenect.grab help file, that frame differencing is as simple as using the 'unique' command?

And again, many thanks to you both. This is exactly the sort of response I was hoping for :)

The idea of frame differencing was something I was trying to calculate to begin with. I knew that mathematically speaking, if you took a base image of an installation space from the web cam, and then compared that with images taken from the web cam at various intervals, then you should be able to compare the matrix's. But I never got around to coding it. I wouldn't say it was laziness this time, but just that I wasn't sure if I could program it accurately enough which is why I started to look into alternatives like cyclops. 

The jit.freenect.grab object looks amazing, and in an email with Rafael Lozano-Hemmer (whom I notice appears in that essay also!) he was recommending the use of kinect's too. They do seem quite powerful for the little you spend on them. 

And again, many thanks to you both. This is exactly the sort of response I was hoping for :)


The wonderful world of cyclops...need a little help.