Automation of detection of transients is one of the things that even pros struggled with for a long time but they seem to have got it pretty well sorted now. I remember back in the late 90s helping Tom Boles with this and the main problem was that you had to accept a pretty high false alarm rate in order to avoid missing stuff. At that time it was just more efficient to blink manually.
I’ve got a large C library of image processing functions written over the years. One of them takes two images at the same scale, cross correlates them in the frequency domain to get alignment, attempts to blur the sharper one to get the same median PSF as the other one, normalises and then subtracts. I have to say that it is not brilliantly effective and that in the end I decided that I had more interesting things to do!
The human eye/brain combination takes some beating.