Case study

Parking ANPR

A desktop app for a gated parking area. It watches a camera feed, detects license plates, reads the plate number, and logs it with the date and time, on its own.

Windows desktop Computer vision Subscription

Parking ANPR main window, showing a live camera preview on the left and a searchable detection log table on the right — The main screen: live camera preview on the left, the day's detection log on the right.

Most gate logs are still a security guard with a notebook, or a camera system that records video nobody has time to watch back. Parking ANPR turns that camera feed into a searchable list instead: every vehicle that passes gets its plate read and timestamped automatically, with a photo kept alongside it.

How it works

Diagram: camera watches the gate, spots a vehicle, reads the number plate, checks if the same car is still there, then saves the record with a photo

Pipeline

camera feed -> plate detection -> text recognition -> duplicate check -> logged + snapshot

A detection model, fine-tuned specifically on license plates rather than a generic object detector, finds each plate in the frame.
The cropped plate is passed to a text-recognition step to read out the plate number.
A vehicle sitting in view for several seconds only gets logged once. A plate already seen in the last cooldown window is matched and skipped, so a shaky reading doesn't create duplicate rows.
Every logged plate lands in a local database with the date, time, confidence score, camera name, and a saved snapshot.
It works with a phone running a free IP-camera app today, and with Hikvision or other network cameras later. Switching is a setting, not a rebuild.

Set up without touching a config file

Camera Settings dialog with camera name, camera type, IP address, port and a Test Camera button — Camera Settings: pick a camera type, fill in its details, test it, then save.

The camera setup window opens on its own the first time the app runs. Pick a camera type (a phone camera app, an existing CCTV recorder, or a direct video address), fill in its details, and click Test Camera to see a live preview before anything is saved. Nobody needs to open or edit a settings file by hand.

Tuned with two sliders, not a spec sheet

Detection Settings dialog with plate detection sensitivity and text reading sensitivity sliders — Detection Settings: two sensitivity sliders cover almost every site.

If a site's lighting or camera angle causes missed plates or false detections, Detection Settings has two sliders for it, one for how eagerly it flags something as a plate, one for how confident a text reading has to be before it's logged. Most installations never need to touch this.

Built to run on the gate PC, not a lab machine

It ships as a standalone Windows installer that bundles everything it needs, including GPU acceleration, so the machine at the gate doesn't need any separate development tools installed. Camera credentials and the logged data stay local to that machine and are never bundled into the installer itself.

Why it's not on GitHub

Unlike the Flutter packages we publish as open source, Parking ANPR is sold directly to customers on a subscription, so there's no public repository or pub.dev listing to link to here. It's still a fair example of the fourth kind of work we do: custom software shaped around a real, specific setup, not a generic template.

Want something like this at your site?

Camera-based automation, a logging system, or anything else purpose-built for how your gate, warehouse or shop floor actually works.

Start a project