Real-time licence plate recognition for smart parking

Staff car parks, delivery yards and restricted zones have the same problem: security wants to know immediately when a vehicle that should not be there enters, but nobody can watch camera feeds all day. Commercial ALPR products are often expensive, closed and hard to connect to the rest of a facility platform.

We needed plate recognition that runs on standard cameras, works with local plate formats, integrates with the permits stored in Smart DigiBuild and raises alerts in seconds.

We built an ALPR service with a classic two-stage computer-vision pipeline:

1. Detection: a YOLOv8 model finds licence plates in each camera frame.
2. Recognition: EasyOCR reads the characters from the cropped plate; post-processing normalises the text and checks it against plate-format rules.
3. Decision: the plate is compared with the list of authorised vehicles.
4. Action: unauthorised vehicles trigger a real-time notification through AWS SNS; every event is stored with its image for review.

A REST API exposes detections, statistics and configuration, so the parking module of Smart DigiBuild can show live events and history.

• Inference service: Python with FastAPI, YOLOv8 for detection, EasyOCR and OpenCV for recognition and image processing.
• Storage: event images in Amazon S3 with lifecycle rules, so images are deleted automatically after the retention period.
• Alerts: Amazon SNS for email and SMS notifications to security staff.
• Monitoring: CloudWatch logs and metrics for throughput, latency and errors.
• Deployment: Docker and Docker Compose, so the same container runs on an edge box next to the camera or in the cloud.
• Privacy: plates are personal data under GDPR; the design includes purpose limitation, short retention and role-based access to images.

The ALPR service became the recognition engine of the Smart DigiBuild parking module. It shows how we approach applied AI: a well-understood model, careful post-processing, clear rules for when to alert a human, and cloud services that make it observable.

Curious how it works in detail? Read ALPR for parking: how plate recognition works, or book an AI sprint for your own computer-vision use case.