Awesome Video . . . I had a setup that probably is very similar to theirs—using 4 3W IR (850nm) power LEDs as IR light source. Did not try to do fiducial detection/recognition, but that seems a good hack/"free time programming” to match what are shown in Video
may i ask you how do you are diffusing the spot created by the LEDs? Are you using some optical approach to do that? Usually the powerfull LEDs are creating very focussed spot…
Cheers,
Sandor
pylin - 18 September 2009 10:10 PM
Awesome Video . . . I had a setup that probably is very similar to theirs—using 4 3W IR (850nm) power LEDs as IR light source. Did not try to do fiducial detection/recognition, but that seems a good hack/"free time programming” to match what are shown in Video
These IR leds, I think, are “uni-directional”, so there is supposedly no focused spot. Anyways, I did not diffuse it, that setup is rear DI with a sanded acrylics as top. With CCV’s ability to do background substraction, it is able to detect blobs. However, it is not very good setup, as you said, blob detection is not very good around those “hot” spots—sanded surface is not even, so some reflection occur.
I used to have quite a few lenses for high power Luxeon leds, I think I can use them on these high power IR leds. Then it would be like shining flashlights on the surface. I will try that when I find these lenses.
as you can see it in the movie, MS is not orienting the leds directly towards the projection surface. More likely the LED is shining horizontally and there is a glass-type plate in the front of the LED. I’m not sure, but i am quessing that the plate is a lens for scattering the light…
pylin - 19 September 2009 09:08 AM
I did not diffuse it, that setup is rear DI with a sanded acrylics as top. With CCV’s ability to do background substraction, it is able to detect blobs. However, it is not very good setup, as you said, blob detection is not very good around those “hot” spots.
I used to have quite a few lenses for high power Luxeon leds, I think I can use them on these high power IR leds. The it would be like shining flashlights on the surface. I will try that when I find these lenses.
Yes, I also thought about using EndLighten acrylics and power IR leds to get evenly diffused IR source. In EndLighten document, they recommended to place light source every foot as poster background. So I think if we put those power IR leds every 6 inches, it should be OK.
Another setup is to use IR led strips, but instead of using expensive EndLighten acrylics, just use some material used for LCD backlight. So the setup would be, place these long strips (in my setup, 24 inches long) every two inches apart, glue them on the same surface as the camera, put the diffusion material on top of these led strips.
I see, it is using adjacent graph tree to recognise fiducial, not feature point detection (which is computationally more expensive). Does Reactivision recognise arbitrary objects? I mean, can we train it to recognize any objects?
as you can see it in the movie, MS is not orienting the leds directly towards the projection surface. More likely the LED is shining horizontally and there is a glass-type plate in the front of the LED. I’m not sure, but i am quessing that the plate is a lens for scattering the light…
The plate is a simple piece of glass and has no special optical properties. The LED is oriented towards the surface and titled slightly outwards. However, it seems to me that Microsoft didn’t show one of the most important parts of the LED assembly in this video, which is a special mirror to focus the light on the four quadrants of the projection surface.
Hmmm, thanks for the info. I believe if I use collimator (lens), I could do the same, but shining from bottom to top. Gee, I got to find those collimators.
