The notification says "Person detected". You open the clip and it is the dog, again, crossing the yard or standing at the back door. After a few days of that, a person alert stops meaning anything, and the one time it really is a person you may not even look.
Person detection in many consumer cameras runs on the camera itself. It has to be small and fast, so it is a classifier that sorts each bit of movement into a short list of fixed labels: person, pet, vehicle, sometimes nothing else. It does not describe the scene. It picks the closest box.
Most of the time the closest box is right. The trouble is the shapes that sit between the boxes:
These cost nothing and are the right first step. Sometimes they are enough on their own.
If the dog still comes through as a person after all of that, the limit is the fixed list of labels, and no slider changes the list.
AI Vision Monitor turns a spare phone, or a camera you already own, into a check you describe in a sentence. On the schedule you set, the app takes a still photograph and a vision model answers your sentence in plain words. There is no fixed list of labels, so the sentence can say what does not count:
The same idea works the other way round, when the dog is the thing you want to hear about:
Spelling out the near misses matters. A loose sentence such as "the dog is on the sofa" can also fire when the dog is on the rug in front of it. Saying what does not count is what keeps the alerts clean.
| Situation | Fixed label camera | A described scenario |
|---|---|---|
| Large dog crosses the lawn | Often "Person" | No alert: an animal, and the sentence says animals do not count |
| Dog stands with its paws on the back door | Often "Person" | No alert: the dog at the door on its own does not count |
| A neighbour walks into the yard | "Person" | Alert: a person is in the back yard |
| Dog climbs onto the sofa while you are out | Motion or "Pet", same as any other movement | Alert, if you wrote the sofa sentence |
| Dog in a coat, seen from behind at dusk | Unreliable | Can still be wrong, see the limits below |
Checks are paid for one at a time. The starter pack is 2000 credits for 2.99 euro, the light model costs half a credit per check and the fast model one credit. New accounts get 50 free credits without a card.
| Schedule | Checks per day | Credits per day (light) | Roughly |
|---|---|---|---|
| Hourly, 08:00 to 20:00 | 12 | 6 | about 30 cents a month |
| Every 30 minutes, round the clock | 48 | 24 | about 1.10 euro a month |
There is no subscription. The cost per day arithmetic for shorter intervals is worked out on its own page.
Why does my camera detect my dog as a person?
Person detection in many consumer cameras is a small classifier that sorts movement into a few fixed labels such as person, pet and vehicle. A large dog, a dog standing up at a door, a dog seen at an angle, half out of frame or in grey night footage can land closer to person than to pet.
What should I try first in my camera's own app?
Turn on pet detection if the camera has it, draw activity zones that leave out the places the dog lies and walks, lower the sensitivity a step, and mount the camera higher so it looks down rather than across. These are free and sometimes enough.
How is a described scenario different from person detection?
Instead of a fixed label, you write a sentence. A vision model looks at a still photograph and answers that sentence, so you can say that a dog or other animal does not count, or ask about the dog on purpose.
Can it tell me when the dog is somewhere it should not be?
Yes, if you write it that way. A sentence such as the dog is lying on the sofa works the same way as a sentence about a person at the door. It does not recognise breeds or tell your dog from the neighbour's.
Does it work inside my existing camera's app?
No. It is a separate app. The usual setup is a spare phone pointed at the same door or yard. A wifi or RTSP camera can be connected through the free companion app for a computer instead.
Will it ever mistake the dog for a person?
It can. No model is right every time, and a dog in a coat seen from behind at dusk can still fool it. Spelling out what does not count in the sentence and testing it on a photograph of your own dog removes most of the confusion.
The general version of this problem is stopping false motion alerts from headlights, rain and shadows. For pets as the subject rather than the noise, see a cat camera from an old phone and leaving two cats alone together. If the opposite is happening and your camera stays quiet when it should not, read why a camera misses events.
Point a spare phone at the yard or the back door and write one sentence that says the dog does not count.
Get it on Google Play Download on the App Store Open in your browser