Machine Vision

cancel
Showing results for 
Search instead for 
Did you mean: 

surveillance program

Hi!

I'm using labview 8.2 and newest Vision library. I'm going to code surveillance program which monitor parking lot and car on it (or selected ROI). Program should show livestream when its running and when someone come insinde that ROI (near car) program takes picture with timestamp and saves it in hard disck.

So question is that what is good approach in this case? I'm quite experienced with Labview but machine vision functions are new to me. So I'm hearing yours advices 🙂

Thanks for advance
0 Kudos
Message 1 of 7
(4,016 Views)
Hi!

That should be pretty easy, well, given that the area around your car isn't changing too much on its own. Compared to standard machine vision applications where you have a great deal of control over the illumination of your object, in surveillance you usually do not have that luxury. Wheather may change (rain, snow, wind moving leaves and twigs etc.). Also the sun is moving its position (yeah, yeah I knows it's actually the earth Smiley Wink).
So usually you'll use a camera with a built in auto-iris to compensate for changes in illumination. Or, if you can spend some more money, you'll use a Thermal/Night Vision Camera.

Anyway, to get back to your question. You could use a region of interest (ROI), or multiple to observe the surrounding of your car. What you are interest in is, if the pixel values are changing significantly within that ROI or the ROIs.
You could use a histogram function and look at the mean value of all pixels within that region, or you could use a pattern match function as well. The pattern match function has a "match count" with a range from 0 to 1000, 1000 being a 100% match. You could use that match count as a threshold for intrusion detection.

There are other ways, but they get more complicated.

We have built many intgelligent surveillance systems.

I hope that helps a bit.

Best,
Markus Tarin

Message 2 of 7
(3,970 Views)
Thank you for your answer and advices. Hope that I'll manage to code that program 🙂
0 Kudos
Message 3 of 7
(3,966 Views)
You can pay me.
I will do it for you.
Its not so easy as you think there are a lot of variables.
you cant write a good application without having knowledge in image processing.
🙂
0 Kudos
Message 4 of 7
(3,956 Views)
Hi!

Feel free to submit your example of yours surveillance program 😉
0 Kudos
Message 5 of 7
(3,941 Views)
Hi,
 
You could try running the cameras quite slowly, like a few frames a second (since people don't move that fast), and find the absolute difference between successive frames.  you could  then threshold the difference image to exclude small variations  and count the number of pixels present in the thresholded image. The pixel count will give an good indication of how much image has changed.  You can convert the ROIs to a mask and mask the images to exclude areas outside the ROI.  
The system should be fairly immune to slow changes in light, but I suspect there would still be issues with rain, snow, trees moving in the wind, etc. The next step would be to use blob analysis to try to reject non-people shaped things.
 
Mike
0 Kudos
Message 6 of 7
(3,935 Views)
Hi Zealous,

I would suggest looking at the pattern matching functions.  The ROI of the area around the car won't change very much once the car is parked, unless someone or something enters the area.  But this method might mean creating a new pattern when the car is moved or if a different car is watched. 

Good luck.
---------------------
Patrick Allen: FunctionalityUnlimited.ca
0 Kudos
Message 7 of 7
(3,855 Views)