Solutions · Video analytics

BrainWhale Vision

Computer vision reads the video stream and turns it into events a person can act on: which vehicle passed, whether the belts were fastened, what the plate was, who walked past the camera, who ran a red light or stopped where it is not allowed, where smoke appeared. Instead of hours of footage the operator gets a list of facts with snapshots.

What it recognises

MODULE 01

Licence plate (ANPR/LPR)

Automatic plate capture on the stream, day and night, in motion and in dense traffic. Body colour, make and model are recognised together with the plate, so a car can be found even when the plate was not read.

MODULE 02

Seatbelt

Detects separately whether the driver and the front passenger are belted. It works through the windshield and stays robust to glare on the glass.

MODULE 03

Face recognition

Finds faces in the frame, matches them against watchlists and raises an alert on a match. Estimates gender and approximate age; the feed shows each person as one row with the best snapshot.

MODULE 04

Red light

Records a vehicle crossing the intersection on a red signal. The violation goes to the log with a snapshot, the camera and the time.

MODULE 05

Restricted zone

Records a vehicle entering a restricted area set on the camera frame. The alert is raised at the moment of entry.

MODULE 06

No stopping

Records a vehicle stopping or parking where it is not allowed. The violation goes to the log with a snapshot, the camera and the time.

MODULE 07

Fire and smoke

Early detection of open flame and smoke in the frame. The alert reaches the operator with a snapshot, the camera and the time.

How it works

The chain is the same for every module: a frame arrives, the detector finds the vehicle, the recognisers read the details, and the result becomes an event with a snapshot.

01
Frame
A frame comes from the camera stream or from the camera itself. Not every frame is analysed — sampling is enough and it saves computing power.
02
Detection
The detector finds vehicles in the frame and cuts out what matters: the plate area, the windshield.
03
Recognition
Each fragment goes to its own model: plate text, belt state, colour, make. A reading below the confidence threshold is marked unknown rather than guessed.
04
Event
The result is written to the database together with the snapshot, the camera and the timestamp, and if a rule matches it raises an alert.
# event example
plate "M BW 2024"
seatbelt driver: no · passenger: yes
colour white
make Toyota
camera cam-07 · 14:32:11
✔ Recognition runs on a graphics card: frames are processed in tens of milliseconds, so a checkpoint camera is handled in real time.

What the operator sees

A web interface is part of the delivery — it opens in a browser and needs nothing installed on the workstation.

Recognition log
All passages with snapshots, filters by plate, date, camera, colour and make.
Map and tracks
Cameras on a map and the path of a specific vehicle between them.
Summary
Counts by day, by camera and by violation — where and when belts are ignored.
Watchlists and alerts
A plate or a face from a list, or an unbelted driver, raises an alert the moment it happens.

Where the events go

Events are not locked inside the analytics: they go to BrainWhale Argus next to the live video and the archive.

Video surveillance system →

See it on your own camera

We connect a module to your camera and show plate and seatbelt recognition in real time.