Software detection of events from allsky timelapse/images

[zief]

Not sure if this is the right category to post this, but it looked close. Let me know if I should post elsewhere.

I've recently set up an allsky camera for recording night skies, and part of the github (software) repository of the person who wrote the instructables info on the project, has scripts that compile a timelapse from all the images taken over a night.

I'm interested in scanning through either the images, or the timelapse video, with software, and automatically labelling events happening in the night. Satellites, meteorites, cloud cover, clarity, aurora, planes, etc. I'm a software engineer, but haven't got any experience with machine learning. I was hoping that there might be some open source software around where I might be able to set up these kind of algorithms, but after Googling for a while, I haven't come up with anything obvious. I was wondering if anyone here had tried anything like this, and might have suggestions of what to read up or what products to use that might help? I believe AWS has Rekognition, but that seems tailored to very specific labelling, and I might have to use SageMaker to train my own models? Just looking for some direction/advice, if anyone on here has that sort of experience!