Publication detail

Geometric Alignment by Deep Learning for Recognition of Challenging License Plates

ŠPAŇHEL, J. SOCHOR, J. JURÁNEK, R. HEROUT, A.

Original Title

Geometric Alignment by Deep Learning for Recognition of Challenging License Plates

English Title

Geometric Alignment by Deep Learning for Recognition of Challenging License Plates

Type

conference paper

Language

en

Original Abstract

In this paper, we explore the problem of license plate recognition in-the-wild (in the meaning of capturing data in unconstrained conditions, taken from arbitrary viewpoints and distances). We propose a method for automatic license plate recognition in-the-wild based on a geometric alignment of license plates as a preceding step for holistic license plate recognition. The alignment is done by a Convolutional Neural Network that estimates control points for rectifying the image and the following rectification step is formulated so that the whole alignment and recognition process can be assembled into one computational graph of a contemporary neural network framework, such as Tensorflow. The experiments show that the use of the aligner helps the recognition considerably: the error rate dropped from 9.6 % to 2.1 % on real-life images of license plates. The experiments also show that the solution is fast - it is capable of real-time processing even on an embedded and low-power platform (Jetson TX2). We collected and annotated a dataset of license plates called CamCar6k, containing 6,064 images with annotated corner points and ground truth texts. We make this dataset publicly available.

English abstract

In this paper, we explore the problem of license plate recognition in-the-wild (in the meaning of capturing data in unconstrained conditions, taken from arbitrary viewpoints and distances). We propose a method for automatic license plate recognition in-the-wild based on a geometric alignment of license plates as a preceding step for holistic license plate recognition. The alignment is done by a Convolutional Neural Network that estimates control points for rectifying the image and the following rectification step is formulated so that the whole alignment and recognition process can be assembled into one computational graph of a contemporary neural network framework, such as Tensorflow. The experiments show that the use of the aligner helps the recognition considerably: the error rate dropped from 9.6 % to 2.1 % on real-life images of license plates. The experiments also show that the solution is fast - it is capable of real-time processing even on an embedded and low-power platform (Jetson TX2). We collected and annotated a dataset of license plates called CamCar6k, containing 6,064 images with annotated corner points and ground truth texts. We make this dataset publicly available.

Keywords

License Plate Recognition, CNN, License Plate Dataset, Image Alignment, Intelligent Transportation Systems

Released

04.11.2018

Publisher

IEEE Intelligent Transportation Systems Society

Location

Lahaina, Maui

ISBN

978-1-72810-321-1

Book

2018 21st International Conference on Intelligent Transportation Systems (ITSC)

Edition

NEUVEDEN

Edition number

NEUVEDEN

Pages from

3524

Pages to

3529

Pages count

6

URL

Documents

BibTex


@inproceedings{BUT155105,
  author="Jakub {Špaňhel} and Jakub {Sochor} and Roman {Juránek} and Adam {Herout}",
  title="Geometric Alignment by Deep Learning for Recognition of Challenging License Plates",
  annote="In this paper, we explore the problem of license
plate recognition in-the-wild (in the meaning of capturing data
in unconstrained conditions, taken from arbitrary viewpoints
and distances). We propose a method for automatic license
plate recognition in-the-wild based on a geometric alignment
of license plates as a preceding step for holistic license plate
recognition. The alignment is done by a Convolutional Neural
Network that estimates control points for rectifying the image
and the following rectification step is formulated so that the
whole alignment and recognition process can be assembled into
one computational graph of a contemporary neural network
framework, such as Tensorflow. The experiments show that the
use of the aligner helps the recognition considerably: the error
rate dropped from 9.6 % to 2.1 % on real-life images of license
plates. The experiments also show that the solution is fast - it
is capable of real-time processing even on an embedded and
low-power platform (Jetson TX2). We collected and annotated
a dataset of license plates called CamCar6k, containing 6,064
images with annotated corner points and ground truth texts.
We make this dataset publicly available.",
  address="IEEE Intelligent Transportation Systems Society",
  booktitle="2018 21st International Conference on Intelligent Transportation Systems (ITSC)",
  chapter="155105",
  doi="10.1109/ITSC.2018.8569259",
  edition="NEUVEDEN",
  howpublished="online",
  institution="IEEE Intelligent Transportation Systems Society",
  number="21",
  year="2018",
  month="november",
  pages="3524--3529",
  publisher="IEEE Intelligent Transportation Systems Society",
  type="conference paper"
}