Number Plate Detection

Automatic License Plate Recognition (ALPR) using YOLOv5 for plate detection and Tesseract OCR for character recognition. This is a college project implemented as a Google Colab notebook.

What it does

1. Detection — a YOLOv5 model is trained to localize license plates in car images (COCO annotations are converted to YOLOv5 label format).
2. Recognition — the detected plate region is passed through PyTesseract (with a Haar cascade for plate-region refinement) to read the plate text.
3. State mapping — Indian RTO codes (e.g. DL, MH, KA) are mapped to their states using Indian RTO.csv.

The pipeline is tested on OpenALPR benchmark images (US, EU and BR sets).

Repository contents

• final_code_vr_iit2018138.ipynb — the full pipeline as a Colab notebook (dataset preparation, YOLOv5 training, detection, OCR).
• Indian RTO.csv — mapping of RTO codes to states.
• Visual Recognition Project.pdf — project report.

Setup

The notebook is designed for Google Colab. Open final_code_vr_iit2018138.ipynb in Colab and run the cells top to bottom. Notable steps:

• Mounts Google Drive and clones the YOLOv5 repository.
• Extracts the car dataset and converts COCO annotations to YOLOv5 labels.
• Trains YOLOv5 (yolov5s, 416×416) on the prepared dataset.
• Runs detection and reads plates with Tesseract OCR.

License

MIT — see LICENSE.