Count cars in the car park with the camera you already have
Nobody knows a car park’s real occupancy. It gets sized by eye, space sales are lost and drivers circle. The camera already watching the car park has the answer in front of it: it just needs reading.
«Your camera is already recording the car park. It just needs someone to read what it sees. That someone need not be a person.»
- Client
- Hotels, shopping centres, private car parks, town councils
- Input
- Existing security camera (RTSP)
- Cadence
- 1 frame every 60 s
- Classes
- Car, van
- Dataset
- 200 – 300 frames of the real scene
- Output
- Real-time occupancy, history, peak hour, free spaces
What you get
Real-time occupancy and free spaces, with history by hour, day and month. Peak hour stops being a hunch and becomes a time series.
The dashboard answers the question that drives investment — does this car park pay off? — and the daily one: raise a capacity alert, open the second floor, send someone to the entrance?
How we do it
One frame every sixty seconds is enough: cars do not move at thirty frames a second. That low cadence removes the need for a GPU in the field, and it is what lets the node be a Raspberry Pi in a weatherproof box.
We draw regions of interest over the bay layout and fine-tune the model on two or three hundred frames of your scene, in sun, in rain and at night. The van that public datasets call a truck becomes whatever you need it to be.
Can be extended with
- LED free-space sign at the entrance
- Integration with the hotel PMS
- Space reservation for guests
- Automatic capacity alerts
Do I need one camera per space?
No. A camera covering the area with enough angle counts dozens of spaces. What matters is not the number of cameras but how many pixels each car occupies in the image.
Does it work at night?
If the camera sees at night, so does the model — provided the dataset includes night frames. That is why we capture across full days before annotating anything.
Does it tell cars from vans?
Yes, if that distinction matters to you. The taxonomy is agreed with you up front: it is cheaper to decide before annotating than after.
Got a camera looking at the car park?
The feasibility study tells you in writing, before you buy anything, what accuracy that specific camera can deliver.
Request a pilot