Camera on a Parked Car That Ignores Cats and Alerts on People

The car sits in an apartment lot that half a dozen stray cats treat as home. A camera that records every movement around the car records cats all night, and a battery camera that does that is flat within a week. The goal is a record of people near the car and silence about the cats.

Three different things called "detection"

Camera listings use the same word for three separate steps, and most of the confusion comes from mixing them up.

• The motion trigger. Most battery cameras sleep until a PIR sensor (passive infrared) notices a warm body moving across its view. That wakes the camera and starts a recording. PIR knows nothing about shapes. A cat, a dog and a person are all warm and moving.
• The person filter. After the camera wakes, a small classifier looks at the video and decides whether to label it person, animal or vehicle. Notifications and saved clips can be limited to the person label.
• A described event. A vision model looks at a photograph and answers a sentence written in plain words. The sentence can say where, which car and what does not count.

Does a person filter work on a head and shoulders?

Usually, yes. Person detection is trained on people seen at every angle and partly hidden behind things, so a head and shoulders over a car roof is a familiar picture. The bigger problem is size. A person standing right next to the car is easy. A person at the far end of the lot may be a few pixels tall in a wide view, and at that size no classifier can tell a person from a bin bag.

The fix is placement. Mount or aim the camera so that someone standing at the car door fills a good share of the frame, instead of fitting the whole lot into the picture.

Why the cats still drain the battery

On a battery camera, the PIR sensor wakes the camera before any person filter runs. The filter can decide not to notify about a cat, but by then the camera has already woken, recorded and spent the energy.

What helps with the battery is reducing how often PIR fires:

• Lower the PIR sensitivity a step or two. A cat is a smaller warm target than a person, and lower sensitivity misses more of them.
• Tilt the sensor up slightly, so the ground and the bonnet where cats walk sit low in its view and a standing person sits in the middle.
• Use activity zones if the camera supports them, to leave out the roof and the space under the car.

More on this kind of tuning is on the page about stopping false motion alerts, and the shape confusion behind labels is covered in camera thinks the dog is a person.

Describing the event in words

AI Vision Monitor works differently from a motion-triggered camera. On the schedule you set, the app takes a still photograph and a vision model answers a sentence you wrote. There is no PIR step, so cats do not wake anything, and there is no fixed list of labels, so the sentence can be specific:

A single photograph cannot measure "more than a few seconds", but checking every few seconds gets close: a person who is still at the door on the next check is someone stopping at the car, not someone passing. Spelling out what does not count matters as much as describing what does. A loose sentence such as "a person near the car" also fires on a neighbour walking to their own car.

When the sentence matches, you get a push notification. An alert can also play a siren or another sound from the phone, turn on its flashlight, or play a short voice line you recorded yourself. A webhook is available in the web version for anything else you want to trigger.

Honest limits

• Snapshots, not continuous video. It checks photographs on an interval, from every few seconds to every few minutes. Someone who walks up, tries the door and leaves between two checks can be missed.
• It needs power and wifi. The usual setup is an old phone used as a camera, on a charger, behind a window that faces the parking spot. A phone is not a weatherproof outdoor camera and should not be strapped to the car.
• Range. A car three floors down and across the lot is small in the photo. A closer window helps more than any setting.
• Night and glass. A phone is not an infrared camera. A dark lot is mostly black in the photograph unless a street lamp lights the car. Glare and reflections from a lit room behind the window can hide the view, so turn the room light off or put the lens against the glass.
• No model is right every time. Test the sentence with a photograph of the real spot before relying on it.

What it costs

Checks are paid for one at a time. The starter pack is 2000 credits for 2.99 euro, and a check costs half a credit on the light model or one credit on the fast model. New accounts get 50 free credits without a card. Checking every few seconds all night uses credits quickly, so limit the schedule to the hours that matter. There is no subscription.

Frequently asked

Will a person filter work if only a head and shoulders show over the car roof?
Usually, yes. Person detection is trained on partly hidden people. Misses come mostly from distance, when a person at the far end of the lot is only a few pixels tall.

Why do cats still drain a battery camera with person detection on?
The PIR motion sensor wakes the camera before the person filter runs. A cat is warm and moving, so it wakes the camera and starts a recording even if no alert is sent.

How do you stop cats from triggering the PIR sensor?
Lower the PIR sensitivity, tilt the sensor so the ground and bonnet sit low in its view, and use activity zones if the camera has them.

How is a described event different from person detection?
Instead of a fixed label, you write a sentence, such as a person standing at the driver's door of the grey car, with cats and people walking past not counting. A vision model answers it from a photograph.

Can an old phone watch the car from a window?
Yes, if it is on a charger, on wifi and has a clear view of the spot. A phone is not a weatherproof outdoor camera, so it belongs indoors behind the glass.

Can it miss someone at the car?
Yes. It checks snapshots on an interval rather than watching continuous video, so a quick pass between two checks can be missed. A dark lot or glare on the window can also hide the car.

Start with one sentence

Put a spare phone on a charger at the window that faces the car and write one sentence about a person at the door, with the cats left out.

Get it on Google Play Download on the App Store Open in your browser