Bangla License Plate Detector with YoLo and Vision Transformers

We have finetuned YoLoV8 on License Plates. YoLo provides the region of interest. We have also trained a document understanding transformer model that can parse bengali texts from a image. Our experimentation has proved that document understanding transformers are mroe powerful compared to OCR methods. We generated 2 million synthetic images of license plates and trained the document understanding transformer with them. Afterwards, we finetuned the model with real world data. We have 3 cameras placed at 3 different angles at the toll plaza. The software individually detects and recognizes a single vehicle from 3 different angles and applies majority voting algorithm to reduce the misprediction probability.

These two Colab notebooks can be explored to understand the training process of License Place Detector and Recognition System.
License Plate Detector
License Plate Recognition System