Forty notifications overnight and every one of them was headlights, rain, a moth on the lens or the neighbour's cat. After a week of that you stop looking at the alerts, which is exactly when the one that mattered arrives. The fix is not a better sensitivity slider. It is telling the camera, in your own words, what actually deserves a notification.
Two things trigger most consumer cameras, and neither of them understands the scene:
The tuning you are offered is a trade, not a solution. Turn sensitivity down and you stop getting alerts for shadows, but you also stop getting them for a person in dark clothing at the far end of the driveway. Activity zones help when the noise sits in one corner of the frame, and they do nothing when the car you want ignored drives through the same patch of driveway as the visitor you want flagged. Person detection on many cameras is either paywalled behind a subscription or limited to a short fixed list of things the vendor decided to support.
AI Vision Monitor runs a vision AI model against the scene and answers the question you wrote. You are not picking from a menu of detectable objects, you are writing a sentence:
Because the model reads the picture rather than counting changed pixels, headlights are headlights and a cat is a cat. It also keeps earlier frames as context, so a car parked in view for three hours is not reported as a new event thirty-six times. The result is that alerts become worth reading again.
With a light AI model (0.5 credits per check) and a check every 5 minutes across a 12-hour night:
144 checks x 0.5 credits = 72 credits, about 0.09 euro per night on the 2.50 euro starter pack.
Longer intervals cost proportionally less, so a check every 15 minutes over the same night is closer to 0.03 euro. New users get 50 free credits, no card required, which is enough to run a full test setup and see what the alerts look like before paying anything. There is no subscription.
Get it on Google Play Download on the App Store
This checks the scene on an interval rather than watching continuously, so something that happens entirely between two checks can be missed. Shorter intervals cost more and catch more, and that trade is yours to set. The device needs internet because the analysis runs server side. A dark, grainy night frame is harder for the AI to read than a lit one, exactly as it is for a person. And this is a notification tool, not a monitored alarm system, so it will not call anyone or dispatch anything on your behalf.
Why does my camera alert on headlights and shadows?
Because standard motion detection reacts to change in the image, not to meaning. Light sweeping a wall changes as many pixels as a person walking does, and the camera has no way to tell the two apart.
Can I get alerts for people only?
Yes, by writing that as the condition, including what to ignore. The model evaluates your sentence against the scene, so cars and animals do not produce an alert unless you asked for them.
Do I have to replace my current camera?
No. Either point a spare phone at the same area, or feed an existing USB, Wi-Fi or RTSP camera in through the free companion app on your computer.
Will it also stop the alerts I do want?
The risk with a sensitivity slider is that it hides real events along with the noise. Here you are narrowing by meaning rather than by threshold, so a person at the door still triggers an alert even when you have told it to ignore every car on the street.
Does it work offline?
No. The camera device needs an internet connection, which is also why an old phone is enough hardware for it.