sayak-2002 / kavach-2023-anpr-system

Automatic Number Plate Recognition and EasyOCR

Step 0: Environment Setup

• conda create -n tfobjdetection pip python=3.9
• conda activate tfobjdetection
• pip install --ignore-installed --upgrade tensorflow==2.11.0

Step 1: Download Kaggle Dataset

• Download Dataset via kaggle API or manually from KAGGLE
• Follow the instructions for Downloading via Kaggle API
• Run kaggle datasets download -d andrewmvd/car-plate-detection in the terminal where the project folder is

Step 2: Move Dataset into a Training and Testing Partition

• Run python ./1.PrepareDataset.py locally

Step 3: Training and Detection

• Upload file ANPR_and_EasyOCR_ColabRun_v1.ipynb in Google Colaboratory
• Remember to Upload archive.tar file of prepared dataset in the step 2 when running ANPR_and_EasyOCR_ColabRun_v1.ipynb in the directory Tensorflow/workspace/images
• Note down the latest checkpoint in the folder Tensorflow\workspace\models\CUSTOM_MODEL_NAME\ e.g. ckpt-100. This will be required to enter in scripts 3.DetectFromImage_EasyOCR.py, 4.DetectFromRealTimeFeed_EasyOCR.py, 5.DetectFromVideos_EasyOCR.py, app.py where LOAD_CHECKPOINT = 'ckpt-101'
• Afer training, Download the compressed file of trained model in the project folder and uncompress it.
• Note: 3.DetectFromImage_EasyOCR.py, 4.DetectFromRealTimeFeed_EasyOCR.py, 5.DetectFromVideos_EasyOCR.py these are optional files.

Step 4: Run the Streamlit UI

• Run streamlit run app.py

Link to Canva Presentation of the Project

• Go Canva