tirsdag, december 12, 2006

Tracking system changed

After much testing we have decided to do a change in the way the system should work. Intensive testing revlealed that our first setup with the infrared lights inside the box had a lot of disadvantages.

  1. It was very hard to get an equal amount of light on the entire plate. It was very hard to get the lamps to sit in the right way. This would render some parts of the plate untrackable.
  2. Infrared light from above would confuse the system. A normal ceiling lamp creates enough infrared light to make the object we track turn into a shadow instead of a lit up dot.
  3. There was a safety risk regarding the use of the many infrared lamps. The human eye cannot see the light and so the reflex that makes us close the eye when the light is too strong does not work. This would not be a huge problem as the screen is a diffusion screen and the lamps are at an angle but nevertheless still a safety risk that would be nice to get rid of.
  4. The system is not very fast as the webcam will have a medium shutter time running at around 20-30 fps or even lower. This is not enough at all for tracking of fast movement.
  5. The system uses a LOT of power as the diodes can handle a forward current of up to 130mA. They are generally used at 100mA but we use them at 20-40mA, closer to the specs of a normal LED. For a useful solution we would need to make it work with a wall plug.
  6. The image quality was in general not good enough for effective tracking. Putting this thing in a club with lasers and powerful lights everywhere would most likely render it completely useless.

So, in general our first solution was very unstable at best. We then decided to once again test out a prototype we made a couple of weeks ago but left behind because we wanted to see if we could make a system that would not require the user to wear anything. The principle is basically the same - the camera will track a shiny dot if infrared light. This solution is based on using some kind of controller, a puck, pen, stick, glove, or even something else with a small battery and a single infrared LED in it. This would be placed on the plate and light down through it straight into the camera. We tested this out and it is by far the favorable solution.

  1. The safety issues are practically gone as long as people don't look straight into the LED which is very unlikely. The LED might even be powerful enough at a level of light that is not hazardous.
  2. The light is very easy to track because it's much more intense than a reflection.
  3. We can bring down the shutter time of the camera to the lowest value possible. First of all, this will filter out anything but the light from the LED. This works because the Vision 2000 screen is a diffusion screen. The diffusion is not very powerful when we have the infrared light pointing straight into it at a few millimeters distance as the light is very intense in a small spot. The light is easily visible beneath the plate. Other lights however, such as lamps above the system will not be seen, as the screen will diffuse the light so much that not enough will enter to be registered by the camera. At the lowest shutter time, even a very bright infrared LED will have to be no more than a couple of centimeters above the plate to be registered by the camera at all. There is absolutely no noise at all and we have a small completely white dot on a completely black background - this doesn't get much better for tracking as we hardly need any filters for noise reduction at all.
  4. The low shutter time has another advantage, namely speed. With the shutter time at the lowest setting, the speed of the camera(fps) go sky high compared to normal shutter times. This is very effective as our application requires as much speed as possible and therefore a nice side effect of the low shutter time.
  5. A lot less power is required as we can power the pointer from a normal 1.5V battery or even a small wrist watch battery.
  6. Because the system is almost completely invulnerable to noise, it will do a lot better in tricky environments such as a club with a lot of lights everywhere.
  7. There is no need to put many lamps inside the box.
  8. The system will be compatible with all sizes of screens as it basically only requires a single infrared LED, a screen and a camera.

The only downside to this system is that the user has to wear something or use something to point with. We don't consider this a huge disadvantage as this is something we have already discussed. A pointer can be a good idea no matter what as you make sure that only the intended user will be able to control the system. A big problem with the old solution was about interference from other things than just things that would confuse the camera. Imagine the DJ is playing and then suddenly somebody throws a peanut or a glowstick onto the plate. How would it react? If users could ruin the DJ's performance just by throwing a peanut onto the table it would most likely be quite a problem. Therefore a dedicated pointer would not be a bad idea as only the DJ would be in control and nothing else could interfere with the system.

We are now working on the new system and will post some pictures soon.

posted by Alex Langberg @ 11:50