Lists

Wednesday, February 26, 2014

Omnivision Supplies 3D Sensors for Google Project Tango

PR Newswire: OmniVision announces that it is working with Google’s Advanced Technology and Projects (ATAP) team to develop tracking, mapping and motion in 3D. OmniVision’s new OV4682 and OV7251 image sensors provide imaging functionality to the project’s Android-based smartphone and development kit.

As the main camera, the OV4682 is the eye of Project Tango’s mobile device. The OV4682 is a 4MP RGB IR image sensor that captures high-resolution images and video as well as IR information, enabling depth analysis. The sensor features a 2um OmniBSI-2 pixel and records 4MP images and video in a 16:9 format at 90fps, with a quarter of the pixels dedicated to capturing IR. From the product page: "The sensor's 2-micron OmniBSI-2 pixel delivers excellent signal-to-noise ratio and IR sensitivity, and offers best-in-class low-light sensitivity with a 40 percent increase in sensitivity compared to the 1.75-micron OmniBSI-2 pixel. The OV4682's unique architecture and pixel optimization bring not only the best IR performance but also best-in-class image quality. Additionally, the sensor reduces system-level power consumption by optimizing RGB and IR timing."

The OV7251 CameraChip sensor is capable of capturing VGA resolution video at 100fps using a global shutter. The low-power VGA sensor provides excellent low-light sensitivity and motion tracking information for device orientation data.

OV4682 Spec

13 comments:

  1. So is it likely that they're extracting the 3D content via a Kinect-style IR projector, analyzed by the sensor's IR channel, or via depth-from-defocus between the R,G,B, and IR channels?

    ReplyDelete
  2. Unlikely. I think they are using a structured light sensor in addition to the RGBI from Omnivision.

    ReplyDelete
  3. Or perhaps they are using a ToF sensor? Capri 3D from PrimeSense?

    ReplyDelete
  4. What kind of IR-sensor is this? In their product brief they are showing something that looks like a thermal camera checking a house for thermal leaks. That confuses me. Or is IR-sensor just a IR-enhanced silicon sensor? Nothing is stated about wavelength sensitivity, as QE curves.
    If they are using structured IR-light, an enhanced IR-sensor will do.

    ReplyDelete
  5. Probably based on their US pat 8.408,821 B2 "Visible and infrared dual mode imaging system".
    Then IR is 950 nm, and they are taking advantage of the water absorbing band blocking sunlight.

    ReplyDelete
  6. Ffor this water vapour bandwidth one has to have an expensive filter on the receiving end and thermo-stabilization on the emitting end. Both are expensive and hard to implement in a tiny form factor. Structured light can not be a solution at 100fps; It might be that they just put more texture on the texture-less areas using IR to patch the holes of conventional stereo. Since they application is indoors only, though, they might do Kinect style IR with random dots everywhere.A third possibility is that to save power they selectively project IR at the area of interest and then assemble a full depth map at a lower fps.

    ReplyDelete
    Replies
    1. why can't structured light be a solution at 100fps?

      Delete
  7. Here's an article with more information; it seems that Movidius supplied the 3D technology. http://www.engadget.com/2014/02/27/smartphone-camera-competition/?ncid=rss_truncated

    ReplyDelete
  8. Movidius only make the processor. The structure light system must be by prinsense or one of the companies copying their tech (apple will squash them later when they make money).

    ReplyDelete
  9. Unless google made a deal with apple... who knows what's going on under the table.

    ReplyDelete
  10. Anybody know what the OV7251 costs?

    ReplyDelete
  11. Any ideas about what lateral and depth resolution can be achieved using any of these technologies?

    ReplyDelete
  12. http://pmdtec.com/news_media/press_release/google_tango_tablet.php

    ReplyDelete

All comments are moderated to avoid spam and personal attacks.