Ok, just a quick update here: the new HD camera is not too bad.
Contrary to what I suspected, the camera started to work right away on one of my Raspberry Pis which still uses an old Summer 2019 “buster” version of the OS (the guys at the Foundation are at a Feb 2020 version already).
My copy of the picamera-lib is a fork with lens-shading build in, otherwise it is not so up-to-date compared to the current picamera-lib.
So in fact, I expected total failure with the new camera.
Well, that was not the case. At least in the resolution I tried, which is 1296x972 pixel, I could stream video from the device using my old software based on the picamera-lib. I could trigger most function that I do not need, for example the image effects. Exposure settings as well as settings for brightness, contrast and saturation (via picamera) were accepted. In video-streaming, I could vary the quality of the MJPEG-encoding over quite some range, effecting as expected the average frame size send to the receiving computer.
I have the impression that the whitebalance algorithm still needs some tuning, but it works sufficiently well in most cases.
What did not work is setting a lens-shading table (well, you would not need that with the HD camera anyway) and activating the capture via the still-port. That might be rather an issue with my software than with the picamera-library. What most probably worked, judging from the slow-down in framerate, is the switch to raw-mode. To be sure about raw mode, I need to write code for reading and decoding the raw image size of the new sensor - that will need some time and effort.
So, in summary, I am surprised how much of the old software interface is available right out of the box for the new HD camera. Here’s an example image to show the color output of the new camera. It was taken with all “autos” on: autoISO, autoExposure and autoColorbalance. Daylight exposure. A 16mm lens @ 1:1.4 setting was used, with an exposure time of 1/30 sec and ISO 130:
(Remember, that is a single video frame, from the videoport, not a still frame (traditionally, the video and still port have different image characteristics, due to different processing pipelines))
Update: I switched resolution now to 2592x1944 pixels, which is an old resolution mode of the v1-sensor. AutoISO is now 183, exposure time 1/64 sec, video streaming at about 7 fps. Here’s a cut-out of the frame
which shows some heavy sharpening happening here. Furthermore, here’s the also enlarged image of a screw drive. Note the staircase-like appearance of bright line visible in head of the bit screw - that should ideally be a single smooth line:
This is the kind of “advanced image processing” you will probably have a hard time to get rid of within the Raspberry Pi software environment. Here’s the full frame were the above cut-outs were taken from: