PS3Eye raw bayer (and 10bit) in Linux - success!

[furrysocks2]

On 30/11/2017 at 17:56, argentum2f said:

Could you condense what you've done into a brief how-to? 

Headline is that if you want to get 10bit/raw, you'll need to be a Linux coder.

 

If you want to use the camera on Windows:

• CLEye driver works out of the box - you get colour only at 8bit
• CLEye SDK lets you write your own application (not a normal driver) - you can get raw bayer at 8bit, which means you can scrape the bayer matrix off the sensor for greyscale and increased sensitivity
• You could look at https://github.com/inspirit/PS3EYEDriver which does a similar thing, but not using the CLEye SDK.

If you want to use the camera on Linux:

• ov534 driver works out of the box - you get colour only at 8bit
• or you can modify the driver as I have done:
  • use register initialisation routines from PS3EyeDriver above
  • adjust frame length and V4L2 pixel format
  • gives you raw bayer 8bit (or you can make it report as grey)
• additionally, if you want 10bit, then additionally set DSP_Ctrl4[0] and adjust frame length and v4l2 parameters appropriately
  • note: nothing I know of will support raw 10bit v4l2 out of the box, yet, so I used oacapture as a platform for changes.

 

That's about it - I'm afraid I don't have time to support in-depth code changes at this time - if you're a programmer the above might help. If not, then perhaps settle for stock. If you need raw bayer because you intend to scrape the bayer matrix off the sensor, then you're limited to Windows and a custom app, and 8bit.

[furrysocks2]

Sensor datasheet refers to "VarioPixel® method for sub-sampling".

I found the following two quotes:

Quote

Software used for increasing light sensitivity and video performance characteristics of image sensor

Quote

Variopixel is a way to get high dynamic range on a sensor by switching from a three transistor pixel to a four transistor pixel for low light conditions and back to three transistors for high light conditions. It is a patented approach to High dynamic range.

 

That sounds interesting but this description seems more akin to "sub-sampling":

Quote

The VarioPixel™ module can decrease the width of the original image by one-half. It supports three modes which selects the either the first, second, or average value of a pair of pixels to compose the new image.

Sounds like horizontal binning (2x1?).

Indeed, for another sensor:

Quote

horizontal VarioPixel (binning) 1:2, 1:3, 1:4, 1:6
vertical VarioPixel® (binning) 1:2, 1:3

 

Also:

Quote

...our VarioPixel technology enables us to enhance the low light video capabilities of our high resolution CameraChips by manipulating multiple pixels to act as a single pixel in order to improve the chip's overall performanec. CameraChips that incorporate this technology can provide significantly improved low light performance at video resolutions giving consumers improved LCD preview capabilities and enhanced video capture. In addition we have produced CameraChips capable of generating usable data in both low light and bright light conditions simultaneously. This high dynamic range technology enables the use of CameraChips in applications generally not suited for image sensors such as in automobiles, security applications and personal identification systems.

 

 

I wonder what registers control that... though perhaps this is simply how it implements 320x240. In which case, that can easily be done later in software, and possibly better (eg ADPB).