Wednesday, April 21, 2010

Omnivision Announces 720p HDR Sensor

PR Newswire: Omnivision announced a new flagship SoC sensor for security - the OV10630. The new sensor combines 720p/30fps HD video with color HDR full image processing pipeline in a 1/2.7 inch optical format. OV10630 delivers 110dB dynamic range in black-and-white and 100dB + in color.

The OV10630 design is based on a 4.2um OmniPixel3-HS pixel having "unparalleled low-light sensitivity". The OV10630 has an active array of 1280 x 800 pixels. The image pipeline has also all required automatic image control functions, including automatic exposure control, automatic white balance, automatic black level calibration, as well as defective pixel correction, gamma correction and lens shading correction.

The OV10630 is now sampling, and is scheduled to enter mass production by the fourth quarter of 2010.

21 comments:

  1. "Unparalleled low-light sensitivity" ! Is Omni using the Invisage marketing guys ?

    ReplyDelete
  2. Omni is such a "hype" company. Sometimes i wonder if they sell hype or image sensors. do you recall how they said wavefront coding would turn 800 years of optics upside-down and we never heard anything about it again?

    ReplyDelete
  3. No Invisage guy comes from OV ! Haha !

    ReplyDelete
  4. well turning a lens upside-down is not a problem, so a 800-old lens is still working !

    ReplyDelete
  5. How does DR in Db relate to DR in f-stops ?

    ReplyDelete
  6. dB = 20log10(x)
    f-stop = log2(x)

    so 1024 => 60dB and 10 f-stop

    110dB => 1: 300000, that means this sensor can take image without accomodation from moonlight (0.2-0.3Lux) to direct Sunshine !!!!!

    Is this possible ?

    ReplyDelete
  7. Sure it is possible, but it does not mean it operates well at 0.2 lux. (Every sensor can operate at 0 lux!). But 110 db is a real capability.

    Anyway, as an example, see the "Lowlight to Sunlight" paper Photobit published in 1999:
    http://ericfossum.com/Publications/Papers/1999%20Program%20A%20Low%20Light%20to%20Sunlight%2060%20fps%2080kpixel%20CMOS%20APS%20camera%20on%20a%20chip.pdf

    Or the more recent Aptina paper presented in 2009 IISW and discussed several times in this blog:
    http://www.imagesensors.org/Past%20Workshops/2009%20Workshop/2009%20Papers/081_Solhusvik_HDR_DCG_final.pdf

    ReplyDelete
  8. Eric,

    Do you talk about the intra-scene DR or inter-scene DR ?

    Thanks !

    ReplyDelete
  9. Every camera can operate with 120dB DR, but this DR is not inside the same scene !!!!

    ReplyDelete
  10. Intrascene dynamic range of course. Interscene DR is not that interesting. Pls. look at Solhusvik's paper as a modern good example of INTRAscene DR > 110 dB.

    By the way, Scott Campbell,suggested a great way to easily verify INTRAscene DR by counting diffraction pattern rings. See:
    "A high dynamic range CMOS APS image sensor"
    http://www.imagesensors.org/Past%20Workshops/2001%20Workshop/2001%20Papers/pg%20137%20YWang.pdf

    ReplyDelete
  11. Not an image scientist so please take it easy on me :)

    I have a few questions for the gallery:

    1) Is the V/Lux-Sec sensitivity stat used by aptina and omni apples to apples against each other?

    2) If they are, is V/Lux-Sec a "complete" sensitivity stat -- or does it not take into account QE and Fill Factor?


    3) Is it possible to compare the "unparalleled sensitivity" of this sensor to a sony ccd (just as a baseline reference) by trying to convert the sensitivity stats in the sony data sheets and comparing it with the V/Lux-Sec stat for omni?
    Here is an example of a sony data sheet for reference where they explain their measurement metodology.
    http://www.sony.net/Products/SC-HP/datasheet/90203/data/a6810506.pdf

    ReplyDelete
  12. by the way, the relevant stats for the sony sensor are on p11 and measurement methodology starts on p13 -- thanks.

    ReplyDelete
  13. most low light surveillance cameras are made with CCD. CMOS is mainly used in IP camera where the exposure time can be long (limited by the net speed :) ).

    Sony EX-View HAD is unparallel in sensitivity !

    Any comments Eric or AT ??

    ReplyDelete
  14. @ 1) Is the V/Lux-Sec sensitivity stat used by aptina and omni apples to apples against each other?

    Pretty much yes, it is.

    @ 2) If they are, is V/Lux-Sec a "complete" sensitivity stat -- or does it not take into account QE and Fill Factor?

    No, V/lux-s is quite incomplete even though QE and FF are included. In particular, it says nothing about color separation quality. Large color crosstalk can increase noise in the final image. In that sense color crosstalk is no less important than high QE. Sometimes QE is sacrificed in the design process to get better color crosstalk. Another thing is color shading, that is what happens in terms of sensitivity and crosstalk across the area. This depends not only on the sensor, but also on the lens used.

    @ 3) Is it possible to compare the "unparalleled sensitivity" of this sensor to a sony ccd (just as a baseline reference) by trying to convert the sensitivity stats in the sony data sheets and comparing it with the V/Lux-Sec stat for omni?

    It's possible albeit very time consuming. Basically, one needs to build a model for the whole camera including full image processing pipeline and optics. Even then one sensor might give better results in the center, while another one is better in the corners. Or one is better in fluorescent illumination, while other is better in incandescent light.

    Not going to spend much time on this analysis, my bet is on Sony, just because the particular CCD you mention has about 4x bigger pixel area, as compared with Omnivision's.

    ReplyDelete
  15. you can make a similar pixel size as in a SONY Ex-View. But can you have the same performance ? The surveillance is a huge market, I wonder why the CMOS guys don't want to make such kind of pixel ?

    It's easy to conclude that SONY Ex-View CCD is far more sensitive than the best CMOS sensors made by Aptina or OV ...

    ReplyDelete
  16. "It's easy to conclude that SONY Ex-View CCD is far more sensitive than the best CMOS sensors made by Aptina or OV ..."

    My question is -- is it easy to conclude the magnitude of any difference in sensitivity by the variables given in the various datasheets? The best I can gather from Image Sensor's reply is the answer is no (thanks for the reply btw Image Sensor).

    If there is an easy way to determine this from the datasheets (no simulations, or actual tests), it would be much appreciated if you could post it. Thanks.

    ReplyDelete
  17. @ The surveillance is a huge market, I wonder why the CMOS guys don't want to make such kind of pixel ?

    I believe both Omnivision and Aptina are overloaded by many projects and chose whatever brings them better profit for each spent R&D dollar. Perhaps they have better options than compete with Sony for this particular price/performance niche on the huge surveillance market.

    Nevertheless, if you represent a big company which needs such a sensor and is ready to pay for it, I'm sure Omnivision and Aptina marketing would be glad to talk about it.

    @ If there is an easy way to determine this from the datasheets (no simulations, or actual tests), it would be much appreciated if you could post it.

    I think the best way is to ask for SNR10 figure. This is quite imperfect figure of merit, but it integrates many factors together, for color imaging, at least. So far I have not seen it appear in datasheets. One of the problems is that SNR10 procedure is not well documented, particularly in the color correction matrix part.

    ReplyDelete
  18. Thanks again Image Sensor.

    BTW omnivision is targeting the security market

    http://www.ovt.com/download_document.php?type=sensor&sensorid=55

    This is a 1/4" sensor (NTSC) with 6um pixels so there is some overlap there...

    ReplyDelete
  19. today's CMOS sensor datasheets are totally useless ! Too much marketing language inside and no enough technical precision. Some puts even misleading information, for example giving > 10V/lux*s but for 680nm wavelength ! A guy not expert with the difference between photometric and radiometric will be misled totally.

    I saw an OV presentation material where they show a comparison between their sensor and a CCD sensor, on this slide, their sensor gives great image but the CCD sensor gives a totally dark image ...

    All the camera comapnies know today that the only way to evaluate a CMOS image sensor is to do their own labo/field test ! No any confidence on the so-called datasheet.

    It's really a pitty !

    ReplyDelete
  20. Which image sensor is good? 1.2 V/lux-sec or with 830mV/lux-sec?

    ReplyDelete
  21. Is image sensor with 830mV/lux-sec is good or with 1.2V/lux-sec is good?

    ReplyDelete

All comments are moderated to avoid spam and personal attacks.