caer VS PySceneDetect

I think it may have. If you look at line 10 of https://github.com/jasmcaus/caer/blob/master/configs.ini, you’ll see that caer has numpy and opencv-contrib-python dependencies that get referenced in its setup.py. If I recall correctly, pip on the nano doesn’t pick up the default numpy and opencv-python system installs, so when you go to install something like caer that has them as dependencies, it will install new copies except the wheel files that it grabs are incompatible. The solution I have found to work is to run something similar to the command above: “pip3 install —no-binary caer —no-binary numpy—no-binary opencv-contrib-python —no-binary typing-extensions —no-binary mypy —force-reinstall caer”. Some of those —no-binary options may not be necessary but they’ll at least ensure pip grabs the source for each of the dependencies and rebuilds it locally rather than using an imcompatible version. This command will take awhile! But you only should have to do it once.

If you mean scene changes, this library works: https://github.com/Breakthrough/PySceneDetect

I saw this library as well https://github.com/Breakthrough/PySceneDetect and I wonder if that might be helpful?

I do this using pyscenedetect. It is a command line tool that also has python bindings you can use so that it can all be scripted. For your use-case in which it is color based scene transitions, you would want to use the content detector. It can not only detect your scene transitions, but can also use a tool like mkvmerge to split the video into individual scenes. I personally use this to detect scene transitions and then use moviepy to annotate over the top of the video using text clips all in a single script.