I work on a paid screenshot api[0] where we have features to either hide these banners and overlays using css, or alternatively we run some javascript to send a click event to what we detect as the 'accept' button in order to dismiss the popups.
It's quite a painful problem and we screenshot many millions of sites a day, our success rate at detecting these is high but still not 100%.
We have gotten quite far with heuristics and are exploring whether we can get better results by training a model.