vargha-kh/plate_detection

License plate detection and OCR

Plate Detection

Introduction

This is a yolov7 model that is fine-tuned to detect plates and recognize the plate from image or video.

Datasets

I used the Iranian plate dataset and added to the roboflow license plate object detection datasets. I used preprocessing and data augmentation such as blur and rotation.

Training

Training notebook (Plate_Detection_Yolov7.ipynb) is in src directory the training result was very good and reached around ~ 0.85 mAP.

Confusion Matrix:

F1_curve:

R_curve:

Whole result

Prediction

I put 11 images in the test set to evaluate the object detection model.

• The car label means plate license

See the test_set folder for more images...

Plate Recognition

After license plate localization, I tested 3 images for cropping and detecting the plates. The results are in plate_inference directory and named by the plate text.

images:

Results:

Usage

For using the plate cropping and ocr you should install the packages in requirements.txt and run by the following command:

python src/yolov7/plate_ocr.py