This repository for project that can Automate Number Plate Recognition (ANPR) in Morocco Licensed Vehicles. πŸ’» + πŸš™ + πŸ‡²πŸ‡¦ = πŸ€– πŸ•΅πŸ»β€β™‚οΈ

Overview

MoroccoAI Data Challenge (Edition #001)

This Reposotory is result of our work in the comepetiton organized by MoroccoAI in the context of the first MoroccoAI Data Challenge. For More Information, check the Kaggle Competetion page !

Automatic Number Plate Recognition (ANPR) in Morocco Licensed Vehicles

In Morocco, the number of registered vehicles doubled between 2000 and 2019. In 2019, a few months before lockdowns due to the Coronavirus Pandemic, 8 road fatalities were recorded per 10 000 registered vehicles. This rate is extremely high when compared with other IRTAD countries. The National Road Safety Agency (NARSA) established the road safety strategy 2017-26 with the main target to reduce the number of road deaths by 50% between 2015 and 2026 [1]. Law enforcement, speed limit enforcement and traffic control are one of most efficient measures taken by the authorities to achieve modern road user safety. Automatic Number Plate Recognition (ANPR) is used by the police around the world for law and speed limit enforcement and traffic control purposes, including to check if a vehicle is registered or licensed. It is also used as a method of cataloguing the movements of traffic by highways agencies. ANPR uses optical character recognition (OCR) to read vehicles’ license plates from images. This is very challenging for many reasons including non-standardized license plate formats, complex image acquisition scenes, camera conditions, environmental conditions, indoor/outdoor or day/night shots, etc. This data-challenge addresses the problem of ANPR in Morocco licensed vehicles. Based on a small training dataset of 450 labeled car images, the participants have to provide models able to accurately recognize the plate numbers of Morocco licensed vehicles.

Table of Contents

Dataset

The dataset is 654 jpg pictures of the front or back of vehicles showing the license plate. They are of different sizes and are mostly cars. The plate license follows Moroccan standard.

For each plate corresponds a string (series of numbers and latin characters) labeled manually. The plate strings could contain a series of numbers and latin letters of different length. Because letters in Morocco license plate standard are Arabic letters, we will consider the following transliteration: a <=> Ψ£, b <=> Ψ¨, j <=> Ψ¬ (jamaa), d <=> Ψ― , h <=> Ω‡ , waw <=> و, w <=> w (newly licensed cars), p <=> Ψ΄ (police), fx <=> Ω‚ Ψ³ (auxiliary forces), far <=> Ω‚ Ω… Ω… (royal army forces), m <=>Ψ§Ω„Ω…ΨΊΨ±Ψ¨, m <=>M. For example:

  • the string β€œ123Ψ¨45” have to be converted to β€œ12345b”,
  • the string β€œ123و4567” to β€œ1234567waw”,
  • the string β€œ12و4567” to β€œ1234567waw”,
  • the string β€œ1234567ww” to β€œ1234567ww”, (remain the same)
  • the string β€œ1234567far” to β€œ1234567Ω‚ Ω… م”,
  • the string β€œ1234567m” to β€œ1234567Ψ§Ω„Ω…ΨΊΨ±Ψ¨",
  • etc.

We offer the plate strings of 450 images (training set). The remaining 204 unlabeled images will be the test set. The participants are asked to provide the plate strings in the test set.
image

Our Approach

Our approach was to use Object Detection to detect plate characters from images. We have chosen to build two models separately instead of using libraries directly like easyOCR or Tesseract due to its weaknesses in handling the variance in the shapes of Moroccan License plates. The first model was trained to detect the licence plate to be then cropped from the original image, which will be then passed into the second model that was trained to detect the characters.

  • Data acquisition and preparation

    First we start by annotating the dataset on our own using a tool called LabelImg. Then we found that the dataset provided by MSDA Lab was publicly available and fits our approach, as they have prepared the annotation in the following form :

    • A folder that contains the Original image and bounding boxes of plates with 2 format Pascal Voc Format and Yolo Darknet Format.
    • And the other folder , contains only the licence plates and the characters bounding boxes with the same formats.
  • Library and Model Architecture

    We have choose faster-rcnn model for both Object detection tasks, using library called detectron2 based on Pytorch and developed by FaceBook AI Research Laboratory (FAIR). A Faster R-CNN object detection network is composed of a feature extraction network which is typically a pretrained CNN, similar to what we had used for its predecessor. This is then followed by two subnetworks which are trainable. The first is a Region Proposal Network (RPN), which is, as its name suggests, used to generate object proposals and the second is used to predict the actual class of the object. So the primary differentiator for Faster R-CNN is the RPN which is inserted after the last convolutional layer. This is trained to produce region proposals directly without the need for any external mechanism like Selective Search. After this we use ROI pooling and an upstream classifier and bounding box regressor similar to Fast R-CNN.

  • Modeling

Training a first Faster-RCNN model only to detect licence plates.

And a second trained separately only to detect characters on cropped images of the licence plates.

The both models were pretrained on the COCO dataset, because we didn’t have enough data, therefor it would only make sense to take the advantage of transfer learning of models that were trained on such a rich dataset.

  • Post-Processing
    Now we have a good model that can detect the majority of the characters in Licence Plates, the work is not done yet, because our model returns the boxes of detected characters, without taking the order in consideration. So we had to do a post-processing algorithm that can return the licence plate characters in the right order.
    1. Split characters based on median of Y-Min of all detected characters boxes, by taking characters where their Y-Max is smaller than Median-Y-Mins into a string called top-characters, and those who have Y-Max greater than Median-Y-Mins will be in bottom_characters.
    2. Order characters in top and bottom list from left to right based on the X_Min of the detected Box of each character.

Owner
SAFOINE EL KHABICH
SAFOINE EL KHABICH
Differentiable architecture search for convolutional and recurrent networks

Differentiable Architecture Search Code accompanying the paper DARTS: Differentiable Architecture Search Hanxiao Liu, Karen Simonyan, Yiming Yang. arX

Hanxiao Liu 3.7k Jan 09, 2023
Rust bindings for the C++ api of PyTorch.

tch-rs Rust bindings for the C++ api of PyTorch. The goal of the tch crate is to provide some thin wrappers around the C++ PyTorch api (a.k.a. libtorc

Laurent Mazare 2.3k Dec 30, 2022
ilpyt: imitation learning library with modular, baseline implementations in Pytorch

ilpyt The imitation learning toolbox (ilpyt) contains modular implementations of common deep imitation learning algorithms in PyTorch, with unified in

The MITRE Corporation 11 Nov 17, 2022
A tensorflow=1.13 implementation of Deconvolutional Networks on Graph Data (NeurIPS 2021)

GDN A tensorflow=1.13 implementation of Deconvolutional Networks on Graph Data (NeurIPS 2021) Abstract In this paper, we consider an inverse problem i

4 Sep 13, 2022
Official code repository for A Simple Long-Tailed Rocognition Baseline via Vision-Language Model.

BALLAD This is the official code repository for A Simple Long-Tailed Rocognition Baseline via Vision-Language Model. Requirements Python3 Pytorch(1.7.

peng gao 42 Nov 26, 2022
Deploy optimized transformer based models on Nvidia Triton server

πŸ€— Hugging Face Transformer submillisecond inference 🀯 and deployment on Nvidia Triton server Yes, you can perfom inference with transformer based mo

Lefebvre Sarrut Services 1.2k Jan 05, 2023
Reproducible research and reusable acyclic workflows in Python. Execute code on HPC systems as if you executed them on your personal computer!

Reproducible research and reusable acyclic workflows in Python. Execute code on HPC systems as if you executed them on your machine! Motivation Would

Joeri Hermans 15 Sep 11, 2022
Source code for our paper "Improving Empathetic Response Generation by Recognizing Emotion Cause in Conversations"

Source code for our paper "Improving Empathetic Response Generation by Recognizing Emotion Cause in Conversations" this repository is maintained by bo

Yuhan Liu 24 Nov 29, 2022
Using pretrained GROVER to extract the atomic fingerprints from molecule

Extracting atomic fingerprints from molecules using pretrained Graph Neural Network models (GROVER).

Xuan Vu Nguyen 1 Jan 28, 2022
State of the Art Neural Networks for Generative Deep Learning

pyradox-generative State of the Art Neural Networks for Generative Deep Learning Table of Contents pyradox-generative Table of Contents Installation U

Ritvik Rastogi 8 Sep 29, 2022
Personal thermal comfort models using digital twins: Preference prediction with BIM-extracted spatial-temporal proximity data from Build2Vec

Personal thermal comfort models using digital twins: Preference prediction with BIM-extracted spatial-temporal proximity data from Build2Vec This repo

Building and Urban Data Science (BUDS) Group 5 Dec 02, 2022
[ICLR 2021] Is Attention Better Than Matrix Decomposition?

Enjoy-Hamburger πŸ” Official implementation of Hamburger, Is Attention Better Than Matrix Decomposition? (ICLR 2021) Under construction. Introduction T

Gsunshine 271 Dec 29, 2022
Books, Presentations, Workshops, Notebook Labs, and Model Zoo for Software Engineers and Data Scientists wanting to learn the TF.Keras Machine Learning framework

Books, Presentations, Workshops, Notebook Labs, and Model Zoo for Software Engineers and Data Scientists wanting to learn the TF.Keras Machine Learning framework

Google Cloud Platform 792 Dec 28, 2022
Classify the disease status of a plant given an image of a passion fruit

Passion Fruit Disease Detection I tried to create an accurate machine learning models capable of localizing and identifying multiple Passion Fruits in

3 Nov 09, 2021
This repository contains the code and models necessary to replicate the results of paper: How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective

Black-Box-Defense This repository contains the code and models necessary to replicate the results of our recent paper: How to Robustify Black-Box ML M

OPTML Group 2 Oct 05, 2022
Weakly Supervised Scene Text Detection using Deep Reinforcement Learning

Weakly Supervised Scene Text Detection using Deep Reinforcement Learning This repository contains the setup for all experiments performed in our Paper

Emanuel Metzenthin 3 Dec 16, 2022
CPU inference engine that delivers unprecedented performance for sparse models

The DeepSparse Engine is a CPU runtime that delivers unprecedented performance by taking advantage of natural sparsity within neural networks to reduce compute required as well as accelerate memory b

Neural Magic 1.2k Jan 09, 2023
External Attention Network

Beyond Self-attention: External Attention using Two Linear Layers for Visual Tasks paper : https://arxiv.org/abs/2105.02358 EAMLP will come soon Jitto

MenghaoGuo 357 Dec 11, 2022
CLNTM - Contrastive Learning for Neural Topic Model

Contrastive Learning for Neural Topic Model This repository contains the impleme

Thong Thanh Nguyen 25 Nov 24, 2022
γ€ŠDeep Single Portrait Image Relighting》(ICCV 2019)

Ratio Image Based Rendering for Deep Single-Image Portrait Relighting [Project Page] This is part of the Deep Portrait Relighting project. If you find

62 Dec 21, 2022