Real-time License Plate Recognition
An intelligent platform designed to instantly recognize and extract text from Bangladeshi vehicle license plates. Whether it's a private car, commercial truck, bike, or CNG, simply upload an image to get accurate, real-time vehicle registration details.
About the Research
This platform is the culmination of a comprehensive research project focusing on Bangladeshi Vehicle License Plate Detection and Character Recognition. Utilizing state-of-the-art computer vision, this system is engineered to solve real-world problems like automated toll collection, traffic law enforcement, smart parking systems, and security surveillance.
The model is trained extensively to detect and recognize both commercial and non-commercial vehicle license plates (including bikes and CNGs), specifically optimized for the Dhaka, Chattogram, Feni, Khulna, and Jashore regions.
Phase 1: Localization
Detection of license plates in unconstrained real-world images using a highly optimized dataset of 1,487 raw images.
- YOLOv11n selected as the superior model
- High Precision: 96.70%
- Overall Accuracy (mAP50): 97.10%
Phase 2: OCR Extraction
Extracting complex strings using a 31-class recognition model trained on 40,094 character instances.
- Detects 31 distinct Bengali and numeric classes
- YOLOv9t selected for string extraction
- High Precision: 94.05%
- Overall Accuracy (mAP50): 94.71%