HI, I been working on interactive installations for public spaces usually i put one camera in the roof pointing to the floor, i always have the problem that i need to have the camera 6 or 7 meters in the roof to track 4 * 4 meters space aproximately and thats a very big limitation since there are a lot of spaces that doesnt allow to put a camera 7 meters up. The thing is that i went to a conference where a guy explained he used 4 cameras to track people in a space instead of just one and you dont need to have a high roof . I would like to know if somebody know this technique for tracking people in spaces with 4 cameras ?
thanks
R.
It seems just: image blending. Each camera sees a portion (quarter) of the floor, and you blend (or stitch) the images, then supply them to the tracking algorithm as single image.
Or... kinect. :D (really depends on what you want, I'm just messing)
On Mon, May 23, 2011 at 6:22 PM, ronni montoya ronni.montoya@gmail.comwrote:
HI, I been working on interactive installations for public spaces usually i put one camera in the roof pointing to the floor, i always have the problem that i need to have the camera 6 or 7 meters in the roof to track 4 * 4 meters space aproximately and thats a very big limitation since there are a lot of spaces that doesnt allow to put a camera 7 meters up. The thing is that i went to a conference where a guy explained he used 4 cameras to track people in a space instead of just one and you dont need to have a high roof . I would like to know if somebody know this technique for tracking people in spaces with 4 cameras ?
thanks
R.
Pd-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-05-23 19:55, Pedro Lopes wrote:
It seems just: image blending. Each camera sees a portion (quarter) of the floor, and you blend (or stitch) the images, then supply them to the tracking algorithm as single image.
computationally more efficient could be to do tracking on the 4 separate images and then "stitch" together the resulting data.
fgmasdr IOhannes
Of course it would. :)
Computationally you could do all in GPU and track everything fast as hell . . . Or parallelize chunks of code into a multicore, and so forth.
But I don't really care. Tracking blobs with OpenCV takes me about 8ms (VGA size camera) in a normal computer, and without a threaded approach, so perfect, suits me without worries, i can have just stich two cams and it gives about 14ms, which is more than great for me.
On May 24, 2011 7:34 AM, "IOhannes m zmoelnig" zmoelnig@iem.at wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 2011-05-23 19:55, Pedro Lopes wrote:
It seems just: image blending. Each camera sees a portion (quarter) of
the
floor, and you blend (or stitch) the images, then supply them to the tracking algorithm as single image.
computationally more efficient could be to do tracking on the 4 separate images and then "stitch" together the resulting data.
fgmasdr IOhannes -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.11 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk3bXlYACgkQkX2Xpv6ydvRmogCfWz4eOymxzS5t3MxgJKuk6IyC QWQAnjR3g94Wv7swKzjUZv6ANDx6XuQT =SPVp -----END PGP SIGNATURE-----
On Mon, 23 May 2011, ronni montoya wrote:
The thing is that i went to a conference where a guy explained he used 4 cameras to track people in a space instead of just one and you dont need to have a high roof . I would like to know if somebody know this technique for tracking people in spaces with 4 cameras ?
That's interesting, because I also have used 4 cameras to track people in a space. It was a long room in which at the middle of the long walls I put two cameras per wall, pointing diagonally towards a car mirror such that each camera would see nearly one half of the space. I used those mirrors to increase the field of view from the standard 53⁰-by-41⁰ to something like 83⁰ wide, which is still not enough, but a big improvement.
You could also use a single mirror and a single camera, if it has enough resolution... but then your mirror will be heavily curved, which means that you wouldn't be able to do the kind of mathematical shortcuts one can take when the mirrors aren't very curvy. And then you have a hole in the middle in which the camera sees itself.
| Mathieu Bouchard ---- tél: +1.514.383.3801 ---- Villeray, Montréal, QC
I once improvised a fish-eye lens with an old dv camera and a glass ball: http://dl.dropbox.com/u/19070857/fisheye_.flv For some other shots, I used a video effect in kdenlive to compensate for the lens distortion: http://dl.dropbox.com/u/19070857/eric-tim-wiel.flv (the flv's should be playable with vlc...) gr, Tim
2011/5/23 Mathieu Bouchard matju@artengine.ca
On Mon, 23 May 2011, ronni montoya wrote:
The thing is that i went to a conference where a guy explained he used 4
cameras to track people in a space instead of just one and you dont need to have a high roof . I would like to know if somebody know this technique for tracking people in spaces with 4 cameras ?
That's interesting, because I also have used 4 cameras to track people in a space. It was a long room in which at the middle of the long walls I put two cameras per wall, pointing diagonally towards a car mirror such that each camera would see nearly one half of the space. I used those mirrors to increase the field of view from the standard 53⁰-by-41⁰ to something like 83⁰ wide, which is still not enough, but a big improvement.
You could also use a single mirror and a single camera, if it has enough resolution... but then your mirror will be heavily curved, which means that you wouldn't be able to do the kind of mathematical shortcuts one can take when the mirrors aren't very curvy. And then you have a hole in the middle in which the camera sees itself.
| Mathieu Bouchard ---- tél: +1.514.383.3801 ---- Villeray, Montréal, QC
Pd-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-
IOhannes m zmoelnig -
Mathieu Bouchard -
Pedro Lopes -
ronni montoya -
tim vets