4 ms·

Inside the Kinect: 4 mics, IR, color, B/W cameras

12 points by gvb 16y ago

joezydeco 16y ago
The system also detects motion by using one infrared camera—classified as Class 1, which indicates that it's not LED-based— and two image sensors, the firm said. The infrared component is a bit more sophisticated than the article implies. Check out this video: http://www.youtube.com/watch?v=r7nRKU0nFxA http://www.youtube.com/watch?v=r7nRKU0nFxA This looks like a LASER projecting a pattern through a hologram or some other kind of interference pattern. The videos of the open-source driver working don't show any of this stuff, unless the Kinect unit already processes this and that's where the depth map comes from.
- gvb 16y ago
  Excellent find! I was really puzzled what a "Class 1 infrared camera" was, but it makes perfect sense to have a Class 1 laser[1] with an infrared camera to pick up the spots. This also makes a lot more sense for detecting distance via triangulation: discrete spots are going to be much easier to triangulate than doing a two-camera vision system, e.g. detect unpredictable edges, correlate the two cameras' edges, and determine distance. The hologram projection concept makes a lot of sense. A full scanning laser would be prohibitively expensive. A hologram projection would be robust and cheap, but less accurate and require much more software processing. [1] http://en.wikipedia.org/wiki/Laser_safety#Class_1 http://en.wikipedia.org/wiki/Laser_safety#Class_1
  - gvb 16y ago
    On further thought, I over-thought that. Parallax calculations aren't necessary. Since the points will be projected at an angle, the distance to the object being painted ("spotted") is going to be directly proportional to the distance between spots.
    - joezydeco 16y ago
      Looking at the datasheet for the PrimeSense reference system that Microsoft used, it does mention a "registration image" and a laser: http://www.primesense.com/?p=514 http://www.primesense.com/?p=514

4 ms·

Inside the Kinect: 4 mics, IR, color, B/W cameras

12 points by gvb 16y ago

joezydeco 16y ago
The system also detects motion by using one infrared camera—classified as Class 1, which indicates that it's not LED-based— and two image sensors, the firm said. The infrared component is a bit more sophisticated than the article implies. Check out this video: http://www.youtube.com/watch?v=r7nRKU0nFxA http://www.youtube.com/watch?v=r7nRKU0nFxA This looks like a LASER projecting a pattern through a hologram or some other kind of interference pattern. The videos of the open-source driver working don't show any of this stuff, unless the Kinect unit already processes this and that's where the depth map comes from.
- gvb 16y ago
  Excellent find! I was really puzzled what a "Class 1 infrared camera" was, but it makes perfect sense to have a Class 1 laser[1] with an infrared camera to pick up the spots. This also makes a lot more sense for detecting distance via triangulation: discrete spots are going to be much easier to triangulate than doing a two-camera vision system, e.g. detect unpredictable edges, correlate the two cameras' edges, and determine distance. The hologram projection concept makes a lot of sense. A full scanning laser would be prohibitively expensive. A hologram projection would be robust and cheap, but less accurate and require much more software processing. [1] http://en.wikipedia.org/wiki/Laser_safety#Class_1 http://en.wikipedia.org/wiki/Laser_safety#Class_1
  - gvb 16y ago
    On further thought, I over-thought that. Parallax calculations aren't necessary. Since the points will be projected at an angle, the distance to the object being painted ("spotted") is going to be directly proportional to the distance between spots.
    - joezydeco 16y ago
      Looking at the datasheet for the PrimeSense reference system that Microsoft used, it does mention a "registration image" and a laser: http://www.primesense.com/?p=514 http://www.primesense.com/?p=514