In the previous comparison I did a week or two ago, the resolution of video and stills output was compared. With respect to reading a car license plate from across the street, there was precious little to choose between the output.
With the latest release of the camera drivers, (
sudo apt-get update && sudo apt-get upgrade) I took the opportunity to run the test again with both old and new drivers.
The most noticeable change is that the field of view of the stills is much greater. It transpired that the previous stills output was upscaled from a cropped part of the sensor, i.e. not all the sensor was being used. It is now, so thank you to James Hughes for the update.
If you click the above images and zoom in to 100%, you’ll see that the car size on both is exactly the same. But, on the stills you can see a lot more of the scene in all directions. This caught me out slightly. I aligned the camera with a test shot using the old drivers (hence a bit of window frame on the left of the ‘new drivers’ still shot).
Here’s the car at 100% (if you click to zoom in the photo) from all four shots and a 400% zoom of the license plate area.
There’s not much to choose between them. You can resolve the car number plate. I still find this impressive for such a tiny lens and sensor. So there is no significant difference in ‘resolution quality’ of output between stills and video. The only difference is that the video output uses a crop from the middle 1920 x 1080 pixels of the sensor. Stills use the full 2592 x 1944.
Screenshot vs FFMPEG
Another thing I did was compare the quality of my “screen-grab from video” with that of individual .jpg frames split from the .h264 file with FFMPEG. The screen-grab technique gives much better quality results. Zooming right in (400%) on the license plate, you can’t resolve the text with the FFMPEGed version.
It may be that there’s an FFMPEG setting for maximum quality, but it looks to me as if my “screengrab” technique is plenty good enough for my needs – and a lot quicker too.
Have fun with your Raspberry Pi camera.
Gallery of shots used in this article.