Tuesday, June 01, 2010

Panasonic Announces D-IMager, a 3D ToF Imager

Business Wire: Panasonic announced the release of its new time-of-flight 3D image sensor, the D-IMager. The sensor is aimed at various gesture control applications, such as out-of-home entertainment, immersive multi-media, interactive digital signage and video arcade games. Panasonic has a web site devoted to the new D-IMager, including a video with many ideas for 3D sensor applications.

One thing I like about D-Imager is that Panasonic is quite open in publishing its spec. The sensor is capable of resolving 160x120 pixels at 20 fps (standard mode) or 30 fps (optional) and is available with a USB2.0 interface. Its viewing angle is 60deg horizontally and 44deg vertically. Its range is from 1.2m to 9m. The sensor is said to be able to work in wide range of lighting conditions, including outdoor at less than 20,000Lux. The depth resolution is from 3cm (max) at 0Lux illumination to 14cm at 20,000Lux. The power is supplied from 18V DC supply with 0.4A avarage current consumption.

The sensor's design is intended to blend with TV displays:

21 comments:

  1. What is the fundamental advantage of ToF compared to classic stereo vision solution ??

    Thanks !

    ReplyDelete
  2. ToF measures distance directly to objects in the scene with little computation required.

    There are many disadvantages of ToF compared to stereo, but stereo is not so great either. In my opinion the advantages and disadvantages depend on the application and scene.

    ReplyDelete
  3. I cannot comment too much on stereo solutions (I haven't worked on them). It's a trade off of having the complexity of the triangulation computation associated with stereo, versus the complexity of requiring an illuminator (driver and source). Some computation is still required to convert photon time of arrival to distance, as well as discriminate between false arrivals (noise) and legitimate return photons. Further, since a ToF imager comes with it's own (non-visible wavelength) illumination source, you don't rely on the ambient light conditions to create the basic images (as req'd for stereo).
    In the ToF world you also need to consider eye (illuminator) safety.

    EF final sentence applies.

    ReplyDelete
  4. I wonder how this might compete with PrimeSense solution as adupted by Microsoft for its Xbox NUI ? price ? avaliability ? size ?

    ReplyDelete
  5. Do you know there is a half watt laser inside PrimeSense stuff too ??

    I'll not buy such kind of stuff. I've worked on IR illumination based face recognition system and I did get some eye problem when I looked too long time before the camera.

    For face recognition, the exposition time is short but for gaming, it could be very long !!!!

    ReplyDelete
  6. I also worked with different 3D Cameras:
    - PrimeSense
    - Optrima
    - Canesta
    Yes, Primesense injuries eyes in case you look on the laser, though they try to reduce it's intensity in such case. But Primesense has 320x240 depth sensor resolution that allows hand tracking on 4 meters, but both Optrima and Canesta has 160x120 depth sensor resolution and hand is too small to be tracked on more than 3 meters on such resolution.
    On other side Optrima (with leds, not lasers) is not light resistant (could be used only for at-home solutions with closed windows to avoid straight sunlight). Primesense is a little more robust, but also has not hight sun light resistance.
    Canesta is the best, but the most closed and documentation pure, and low depth sensor resolution (160x120) does not allow track hands of people on more than 3 meters.
    I didn't tried Panasonic, but it seems it would be very similar to Canesta, but more raw.

    ReplyDelete
  7. By the way real TOF cameras such as Canesta gives much better data then any stereo solutions and much more resistant to environment conditions

    ReplyDelete
  8. Dear Artem,

    The frame rate could be another issue. I saw some of TOF cameras such Mesa Imaging, PMD, etc... The raw data is very noisy and temporal smoothing is needed. This smoothing operation reduces a lot the frame rate. I've difficulty to image the hand tracking at 3m with 25fps !!!

    What is your opinion ?

    ReplyDelete
  9. I think that Canesta is demonstrating 320x240 version actually ....

    ReplyDelete
  10. This comment has been removed by the author.

    ReplyDelete
  11. Dear, Anonymous,
    hand tracking on TOF on 25 fps more than possible.
    I didn't see Mesa Imaging, PMD, etc., but with Canesta even on 160x120 depth resolution sensor it's OK. By the way fps on this camera can be rised up to 60.
    I heard that Canesta announced about 320x240 version, but I didn't worked with such sensor.

    ReplyDelete
  12. >>> The raw data is very noisy and temporal smoothing is needed
    Yes, you need preprocessing filtration to process usefull data from depth sensor, but anyway it's easier and much reliable to recognize hand gestures using depth mask than color 2D image only, because the main problem of commercial level hand-tracking applications on 2D color camera is background noise (objects or other people on background).
    As for stereocameras I didn't try professional stereocameras, but as for two simple calibrated web-cameras then there are a lot of issues and not much advantages comparing to single 2D cameras for tracking / gesture recognition.

    ReplyDelete
  13. How about the use of optic flow ?
    http://www.youtube.com/watch?v=ssINeWRb58M&feature=PlayList&p=296F3619D33C4179&playnext_from=PL&index=6

    ReplyDelete
  14. >>> How about the use of optic flow ?
    Raw optical flow can be used only to make some drafts for yourself.
    How about noise on background/foreground, how about tracking control user in the crowd, how about distinguishing hands (left/right) and other parts of the body and other objects that are not control user?
    Hey, you are talking as a kid or about things that can be done for a few hours. TOF is not required for such child jobbing.
    I'm talking about COMMERCIAL SOLUTIONS, otherwise TOF is not required,

    ReplyDelete
  15. Dear Artem,

    TOF is not a silver ballet neither ! You will face the same problem as other computer vision based solution. I can find all the cases that TOF cannot work.

    The key point is what is needed for gesture control ???

    ReplyDelete
  16. You are absolutely right, the best solution is usage of hybrid data: depth data plus color image.

    ReplyDelete
  17. As for Canesta, it's better than Optrima for our goals, but it has also many disadvantages. Noise is too hight to use only depth mask of Canesta to make something valuable on significant distance (more than 2-3 meters)...

    ReplyDelete
  18. PrimeSense solution offers both depth and color image and sound as well.

    ReplyDelete
  19. Hello Sir,
    I want 3D camera for 3D gesture recognition (with deph information). Please tell me its cost and how will I get it?
    Email ID: sonal.kumari1910@gmail.com

    Sonal Kumari

    ReplyDelete
  20. There are a lot more companies on the market. Take a look at PMDTechnologies. You can find a whitepaper about the comparsion of stereo and TOF on the webpage.

    ReplyDelete
  21. Panasonic has copied the MESA approach! I think there will be some ToF patent problems in the future. From my point of view, MESA and PMD have the most established patents worldwide!

    ReplyDelete

All comments are moderated to avoid spam.