This project aims to detect and recognize the regular Tunisian license plates with high accuracy using Mask-RCNN and Image processing techniques. It is part of the Computer Vision for License Plate Recognition Challenge from Zindi (https://zindi.africa/competitions/ai-hack-tunisia-2-computer-vision-challenge-2) designed specifically for the AI Tunisia Hack 2019. The data is composed of two datasets:
- A set of vehicle images (900 images) taken from the internet and annotated manually. The annotations are the coordinates of the bounding box containing the license plate.
- A set of license plate images (900 images) where the annotations are the text written in the license plate.
1- Create conda environment for project with python version 3
conda create -n "name" python=3
2- Install openCV, pandas, numpy, matplotlib and jupyter
command for jupyter:
conda install -c conda-forge jupyterlab
command for openCV:
conda install -c conda-forge opencv
commands for the rest:
conda install pandas, numpy
command for matplotlib:
conda install -c conda-forge matplotlib
command for pillow:
conda install -c anaconda pillow
command for keras:
conda install -c conda-forge keras
conda install -c anaconda tensorflow-gpu
conda install -c anaconda cudatoolkit
3- Launch jupyter notebook and start exploring the data
This model works fine with tensorflow=1.14 and keras=2.2.5
This project still under development
This project is free to explore, contribute and may be redistributed under the terms specified in the LICENSE file.