My neighbor's Ring camera flagged a raccoon as a "suspicious person" and now I don't trust AI labels
Last week my neighbor in Cleveland got a push alert from his Ring camera that said "person detected" at 2 AM, so he grabbed a flashlight and went outside. Turned out to be a very fat raccoon tipping over his trash can, and the AI labeled it as a person on the thumbnail. I've been messing with edge detection models on a Raspberry Pi for a month now and I get why it happens, the shape and heat signature can look close enough at night with bad lighting. Still, it got me thinking about how many "smart" alerts are just guessing with extra steps. Has anyone found a camera or open source setup that doesn't cry wolf every time a cat walks by the porch?