Automatic No Plate Recognition

Tech Stack

The Problem

• Manual plate identification at checkpoints caused traffic congestion during peak hours.
• Human operators struggled to reliably read plates under low light, rain, and partial occlusion.
• No real-time system could cross-reference plates against vehicle databases at motorway speeds.
• High miss rates on damaged or non-standard plates created security and compliance gaps.
• Manual logging was inconsistent and error-prone, compromising audit trail integrity in regulated environments.

Our Solution

• Developed a YOLOv9-based detection pipeline achieving real-time plate localisation across video streams.
• Implemented OCR post-processing optimised for varied fonts, angles, and international plate formats.
• Built a multi-condition training dataset covering night, wet weather, and partial occlusion scenarios.
• Designed a database integration API enabling instant real-time plate cross-referencing at scale.
• Optimised the inference pipeline to process multiple simultaneous video feeds on standard hardware.

Impact

Enabled automated vehicle identification at scale, processing hundreds of plates per minute with 97% accuracy.