
i logged every camera alert for 3 months. 76% was noise
ran the numbers on my two reolinks after filtering everything for about 3 months.
20,829 motion events total across both cameras.
front door: person 2,256 · animal 676 · vehicle 757 · nothing 7,112
back door: person 721 · animal 646 · vehicle 0 · nothing 8,661
so out of ~20,800 events, around 5,050 actually contained something worth knowing about.
the other 76% was wind, shadows, light changes, insects, the usual.
alarm delay and detection zones help trim some of it — someone pointed that out on my last post and they're right.
but they never touched the volume the way filtering on the snapshot level did.
the wild part isn't the person/vehicle counts.
it's the "nothing" pile.
most of every alert my cameras ever sent was basically a tree moving, a cloud passing, changing light, or something flying close to the lens.
i ended up building a filter that does this automatically — it checks the snapshot before anything hits my phone and only forwards alerts that actually contain a person, vehicle, or animal.
it's my own project, guardian.camera, if anyone else is dealing with the same thing.