This repository is the implementation of Vehicle and License Plate Recognition with Novel Dataset for Toll Collection (VT-LPR).
The trained models can be downloaded from Google Drive
Tolling efficiency in manual toll collection system is low and time consuming, which results in delayed traffic and long queue of vehicles. This also requires human effort and resources.
- Python>=3.6
- Streamlit==1.0.0
- PyQt5==5.15.4
- PyQt5Designer==5.14.1
- OpenCV==4.5.1.48
- pillow
- matplotlib
- pandas
- numpy
- tensorflow
- tflite
Note: These packages can be installed through Pip package installer. This code is tested with these versions.
Image based vehicle type and license plate recognition include many challenges such as:
- Intra-class variations
- Background clutter
- Non-uniform position
- Multiple license plates
- Non-uniform fonts
- Partial occlusion
- Damaged license plates
- Lack of license plates
We divide the Vehicle type and license plate recognition task into three steps: Vehicle type recognition, License plate localization and License plate character recognition. So we prepared three dataset named: preprocessed dataset for Vehicle type recognition, lp-detect dataset for license plate location and lp-read dataset for license plate character recognition. As we're training separate model for each task, all the vehicle images are annotated accordingly as shown below.
The proposed framework is aim to automatically collect toll amount by end-to-end image-based vehicle type and license plate recognition. Input to the model is an image and the outputs are recognized vehicle type and license plate character. When an image is fed to the first network called VT-Net, it recognize the vehicle type and crop the detection results. The cropped image is then fed to the secong network called LP-Net, which localize and crop the license plate. Finally, the last model called CR-Net is applied on the cropped license plate image to recognize the characters in the license plate.
- Download and extract trained model weights from Google Drive.
https://drive.google.com/file/d/1nGe7kQaKwKzzrvPYSd41ScuSuj2D33-C/view?usp=sharing
- Clone this repository
!git clone https://github.com/usama-x930/VT-LPR.git
- Put model weights in the downloaded folder.
- Run the Streamlit code file using:
streamlit run streamlit_code.py
- Run PyQt5 GUI using this command.
python vt_lpr_gui.py
