For dataset with annotations use this link :- https://www.kaggle.com/datasets/andrewmvd/car-plate-detection

License Plate Detection

This project detects license plates in images using a custom trained TensorFlow object detection model.

Overview

• Uses TensorFlow Object Detection API to train a custom SSD model on license plate data
• Model is trained on annotated images of cars with license plates
• Produces bounding boxes around license plates in new images
• Applies OCR on the detected ROI to read the license plate text

Model Architecture

The model architecture is SSD MobileNet V2 FPNLite pretrained on COCO dataset and fine-tuned on custom license plate data.

Usage

number_plate_detection.ipynb - Main notebook to view demo predictions on test images.

The trained model files are in /Tensorflow/workspace/models/my_ssd_mobnet

To use the model on new images:

• Load model pipeline config and build detection model
• Restore checkpoint
• Run inference using detect_fn() on input image
• Filter detections by threshold and apply OCR

References

• TensorFlow 2 Object Detection API
• Papers/tutorials referred for implementation