egyptian-license-plate-detector

Egyptian License Plate Detector

This is a YOLOv8n model trained to detect Egyptian license plates, converted to ONNX format.

Model Details:

• Framework: YOLOv8n (converted to ONNX)
• Dataset: EALPR Egyptian Vehicles Dataset
• Metrics (Validation Set):
  • mAP@50: 99.5%
  • mAP@50-95: 91.8%

Usage: This model can be used for real-time detection of Egyptian license plates in images or video streams. It is intended to be part of a larger system that includes an OCR component for reading the plate characters.

Files:

• best.onnx: The ONNX converted YOLOv8n model.
• config.json: Configuration file with model metadata and labels.

Example Python Usage (with onnxruntime and easyocr for OCR):

import onnxruntime as ort
import numpy as np
import cv2
import easyocr

def preprocess_image(image_path, img_size=(640, 640)):
    img = cv2.imread(image_path)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    img_resized = cv2.resize(img, img_size)
    img_normalized = img_resized.astype(np.float32) / 255.0
    img_transposed = np.transpose(img_normalized, (2, 0, 1)) # HWC to CHW
    img_batch = np.expand_dims(img_transposed, axis=0) # Add batch dimension
    return img, img_batch

def postprocess_output(output, original_img_shape, img_size=(640, 640), conf_threshold=0.25, iou_threshold=0.7):
    # output format is (batch_size, 5 + num_classes, num_boxes)
    predictions = np.squeeze(output).T # Transpose to (num_boxes, 5 + num_classes)
    
    # Filter out low confidence detections
    scores = np.max(predictions[:, 4:], axis=1)
    predictions = predictions[scores > conf_threshold]
    scores = scores[scores > conf_threshold]

    if len(predictions) == 0:
        return []

    # Get class IDs and bounding boxes
    class_ids = np.argmax(predictions[:, 4:], axis=1)
    boxes = predictions[:, :4]

    # Convert boxes from normalized YOLO format to pixel coordinates
    # x_center, y_center, width, height
    x_center, y_center, width, height = boxes[:, 0], boxes[:, 1], boxes[:, 2], boxes[:, 3]
    x1 = (x_center - width / 2) * img_size[0]
    y1 = (y_center - height / 2) * img_size[1]
    x2 = (x_center + width / 2) * img_size[0]
    y2 = (y_center + height / 2) * img_size[1]

    # Scale bounding boxes to original image size
    ratio_w = original_img_shape[1] / img_size[0]
    ratio_h = original_img_shape[0] / img_size[1]
    
    x1 *= ratio_w
    y1 *= ratio_h
    x2 *= ratio_w
    y2 *= ratio_h

    bboxes = np.column_stack([x1, y1, x2, y2]).astype(int)

    # Apply Non-Maximum Suppression
    indices = cv2.dnn.NMSBoxes(bboxes.tolist(), scores.tolist(), conf_threshold, iou_threshold)
    if len(indices) > 0:
        indices = indices.flatten()
        return [{'box': bboxes[i], 'score': scores[i], 'class_id': class_ids[i]} for i in indices]
    return []

# Load ONNX model
session = ort.InferenceSession("best.onnx")
input_name = session.get_inputs()[0].name
output_name = session.get_outputs()[0].name

# Initialize EasyOCR reader
reader = easyocr.Reader(['ar', 'en'], gpu=True) # Set gpu=False if no GPU available

# Example usage
image_path = "path/to/your/image.jpg"
original_img, preprocessed_img = preprocess_image(image_path)

# Run inference
outputs = session.run([output_name], {input_name: preprocessed_img})

# Post-process and get detections
detector_results = postprocess_output(outputs[0], original_img.shape)

if detector_results:
    for res in detector_results:
        x1, y1, x2, y2 = res['box']
        plate_crop = original_img[y1:y2, x1:x2]
        
        # OCR on the cropped plate
        ocr_result = reader.readtext(plate_crop, detail=0)
        plate_text = " ".join(ocr_result)
        print(f"Detected plate: {plate_text} (Confidence: {res['score']:.2f})")

        # Visualize (optional)
        img_with_box = original_img.copy()
        cv2.rectangle(img_with_box, (x1, y1), (x2, y2), (0, 255, 0), 2)
        cv2.putText(img_with_box, plate_text, (x1, y1 - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
        # plt.imshow(img_with_box)
        # plt.title(plate_text)
        # plt.axis('off')
        # plt.show()
else:
    print("No license plate detected.")