CarPlateDetection

A YOLOv5-lite based program using a custom model fine-tuned to detect French car plates, with added Optical Character Recognition (OCR) to read plates characters.

Credits

This project is based on YoloV5-lite.

Demonstration

Here is a short video demonstrating the program detecting and reading a french licence plate:

CarPlateDetection.webm

Requirements

• Python 3.10
• pip

Installation

1. Clone the repository:

git clone https://github.com/Toufik1247/CarPlateDetection.git

2. Install Tesseract-OCR

sudo apt update
sudo apt install tesseract-ocr

3. If python3-venv is not installed, install with

sudo apt-get install python3-venv

4. Create your virtual environment:

python3 -m venv venvcarplatedetection

5. Activate your virtual environment:

source venvcarplatedetection/bin/activate

6. Install dependencies:

cd CarPlateDetection
pip install -r requirements.txt

7. Open upsampling.py in file editor:

cd ..
nano venvcarplatedetection/lib/python3.10/site-packages/torch/nn/modules/upsampling.py

8. Replace the following function:

def forward(self, input: Tensor) -> Tensor:
    return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners,
                         recompute_scale_factor=self.recompute_scale_factor)

with (be careful to indent correctly)

def forward(self, input: Tensor) -> Tensor:
    return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners,
                         # recompute_scale_factor=self.recompute_scale_factor
                         )

Usage

Run the program using your webcam:

cd CarPlateDetection
python3 detect.py --source 0