License Plate Recognition System

⚠️ Status: Under Construction / Refactoring

This project is currently undergoing a major refactoring and optimization process. The codebase is being updated to improve detection accuracy and robustness, particularly for double-row license plates.

Overview

This project aims to implement a robust license plate recognition system. The current focus is on the Plate Localization module, which has been significantly enhanced to handle various lighting conditions and plate types.

Core Logic: Plate Localization (plate_detector.py)

The plate localization algorithm has been redesigned with the following pipeline:

1. Center Crop: The image is cropped to focus on the central region, reducing processing time and background noise.
2. Dual-Direction Sobel Edge Detection:
   • Combines Sobel X (vertical edges) and Sobel Y (horizontal edges) with equal weights.
   • This ensures that both the vertical strokes of digits/letters and the horizontal strokes of Chinese characters are captured.
3. Pre-Dilation:
   • Applies a small (3x3) dilation to the binary edge image.
   • This strengthens weak edges, ensuring they survive the subsequent morphological operations.
4. Morphological Closing:
   • Uses a small (3x3) kernel to connect broken plate borders.
   • Avoids using large kernels that might merge the plate with surrounding background clutter.
5. Contour Analysis (RETR_TREE):
   • Extracts contours using the RETR_TREE mode to detect nested structures.
   • Leverages the hierarchy to find "parent" contours (plate borders) that contain "child" contours (characters).
6. Candidate Scoring:
   • Candidates are scored based on Aspect Ratio (AR) and Distance from Center.
   • High Weight on AR: The system strongly prioritizes candidates with an AR close to 1.98 (standard for double-row plates).
7. Projection Refinement:
   • The detected region undergoes a projection analysis (horizontal and vertical) to precisely crop the plate boundaries, removing excess padding.

Getting Started

Prerequisites

Install the required dependencies:

pip install -r requirements.txt

Running Tests

To verify the plate localization logic:

python 01_Test_locate_plate.py

TODO

• Refactor Plate Localization (Dual Sobel, AR Scoring)
• Optimize Character Segmentation
• Update Character Recognition Logic
• End-to-end Pipeline Testing

---

The rest of this documentation is intentionally left blank as the project evolves.