Automatic License Plate Recognition (ALPR)

The Python script is designed for real-time vehicle detection and license plate recognition, utilizing YOLOv11. It features a three-layer cascading model that performs forward predictions to extract license plate information. The application can take input from a video stream or video file facilitated through the OpenCV library.

List of contents :

• Automatic License Plate Recognition (ALPR)
  • Usage
  • Function Params
  • Testing
    • 1. Setup
    • 2. Change Arguments Input
    • 3. Running
    • 4. Result

Usage

1. Install required depedencies

   pip install torch opencv-python pandas ultralytics

2. Make sure you already setup arguments for run_model() function in run.py script

   # Example `run.py`
   
   if __name__ == '__main__':
      run_model("videos/test.mp4", 
        skip_frames=6,
        write_results=True,
        write_interval=3,
        export="videos/output/test_output.mp4"
      )

3. Hit and run

   python run.py

4. The OpenCV window will poup-up on your device screen and start playing your video input
5. Enjoyy !!

Function Params

Below is a list of parameters that run_model() function receives:

• video_path: str - The input video file path location.

• skip_frames: int - By default, the application always predicts every single frame of video pics, to reduce computing resources you can change this parameter value.

• width_resize: int|None - Resize application window width while maintaining aspect ratio from video input.

• display_frame: bool - Show OpenCV window frame when application playing video input.

• write_results: bool - Output the result of predictions into files containing crop images of the vehicle, license plate, and character recognition stored in the "detections" folder.

• write_interval: int - The seconds interval result of predictions output will generated.

• export: str|None - Output file path of streamed video prediction result.

Testing

Application accuracy is evaluated through image-level testing; if the application can correctly predict the full character results from license plates, then the results are considered accurate.

1. Setup

Create new directory and give label on the image with expected result of license plate information. Example: A D 1 2 3 C D.jpg

The directory tree will look like this :

images
├── A A 1 0 3 4 F X.jpg
├── A B 1 5 4 0 I J.jpg
├── A B 4 9 1 6 D A.jpg
├── A D 2 A.jpeg
├── A D 4 4 2 6 L B.jpg
└── W 4 3 9 7 B U.jpg

2. Change Arguments Input

After labeling the image open test.py file then make sure test() functions arguments input pointed into images directory.

# Example `test.py`

if __name__ == '__main__':
    test("images")

3. Running

Just run test.py to start accuracy test and wait till finish.

python test.py

4. Result

When the test result is finish application will give output like shown below :

Total images   : 20
Accuracy (num) : 14/20
Accuracy ($%)  : 70.00%

On the other side, you can also review test results by opening test-output.csv files or viewing output directory that is generated inside the image test directory.

├── images
│   └── output
│       ├── correct
│       └── incorrect
└── test-output.csv