Project Overview

Developed a license plate detection and recognition system using YOLOv8. A deep learning project focused on computer vision.

Technologies Used

• Python
• YOLOv8 (You Only Look Once, version 8)
• OpenCV
• Pytesseract (for OCR)
• Ultralytics YOLO
• Pandas for data manipulation
• Matplotlib for visualization

Dataset

Custom dataset with 225 annotated images of vehicles.

Methodology

1. Data Preparation:
   • Converted XML annotations to YOLO format
   • Normalized bounding box coordinates
2. Model Training:
   • Trained YOLOv8 nano and small models for 100 epochs each
3. Model Evaluation:
   • Analyzed training metrics and visualized results
4. Inference Pipeline:
   • Developed custom functions for detection, NMS, and OCR

Results

Training results for YOLOv8 small model:

Predictions on validation set:

Additional Performance Metrics

F1 Score Curve:

Precision Curve:

Result on test set:

Key Achievements

• Successfully trained custom object detection models for license plates
• Implemented an end-to-end pipeline from detection to OCR
• Worked with real-world, unstructured data

Future Improvements

• Expand dataset size and diversity
• Experiment with more advanced YOLO variants
• Implement real-time detection for video streams
• Improve OCR accuracy for challenging plates