Road sign detection and recognition using fuzzy artmap
a case study Swedish speed-limit signs
Document identifier: oai:dalea.du.se:2253
Keyword: Traffic signs,
Color segmentation,
Outdoor images,
Fuzzy ARTMAP,
Classification.Publication year: 2006Abstract: In this paper, a novel approach is developed using Fuzzy ARTMAP Neural Networks to recognize and classify Swedish road and traffic signs. The Swedish Speed-Limit signs are selected as a case study, but the system can be applied to other signs. A new color detection and segmentation algorithm is presented in which the effects of shadows and highlights are eliminated. Images are taken by a digital camera mounted in a car. Segmented images are created by converting RGB images into HSV color space and applying the shadow-highlight invariant method. The method is tested on hundreds of outdoor images under shadow and highlight conditions, and it shows high robustness; in 95% of cases of correct segmentation is achieved.
Classification is carried out by two stages of Fuzzy ARTMAP which are trained by 210 and 150 images, respectively. The first stage determines the border of the sign and the second stage determines the pictogram. Training and testing of both stages are made offline, using still images. In online mode, the system loads the Fuzzy ARTMAP and performs recognition process. An accuracy of 96.7% is achieved in Speed-Limit recognition and more than 90% as whole accuracy.
Authors
Hasan Fleyeh
Högskolan Dalarna; Datateknik
Other publications
>>
Syed Gilani
Other publications
>>
Mark Dougherty
Högskolan Dalarna; Datateknik
Other publications
>>
Record metadata
Click to view metadata
header:
identifier: oai:dalea.du.se:2253
datestamp: 2021-04-15T12:25:47Z
setSpec: SwePub-du
metadata:
mods:
@attributes:
version: 3.7
recordInfo:
recordContentSource: du
recordCreationDate: 2006-08-16
identifier: http://urn.kb.se/resolve?urn=urn:nbn:se:du-2253
titleInfo:
@attributes:
lang: eng
title: Road sign detection and recognition using fuzzy artmap
subTitle: a case study Swedish speed-limit signs
abstract: In this paper a novel approach is developed using Fuzzy ARTMAP Neural Networks to recognize and classify Swedish road and traffic signs. The Swedish Speed-Limit signs are selected as a case study but the system can be applied to other signs. A new color detection and segmentation algorithm is presented in which the effects of shadows and highlights are eliminated. Images are taken by a digital camera mounted in a car. Segmented images are created by converting RGB images into HSV color space and applying the shadow-highlight invariant method. The method is tested on hundreds of outdoor images under shadow and highlight conditions and it shows high robustness; in 95% of cases of correct segmentation is achieved. \nClassification is carried out by two stages of Fuzzy ARTMAP which are trained by 210 and 150 images respectively. The first stage determines the border of the sign and the second stage determines the pictogram. Training and testing of both stages are made offline using still images. In online mode the system loads the Fuzzy ARTMAP and performs recognition process. An accuracy of 96.7% is achieved in Speed-Limit recognition and more than 90% as whole accuracy.
subject:
@attributes:
lang: eng
topic: Traffic signs
@attributes:
lang: eng
topic: Color segmentation
@attributes:
lang: eng
topic: Outdoor images
@attributes:
lang: eng
topic: Fuzzy ARTMAP
@attributes:
lang: eng
topic: Classification.
language:
languageTerm: eng
genre:
conference/other
ref
note:
Published
3
name:
@attributes:
type: personal
authority: du
namePart:
Fleyeh
Hasan
role:
roleTerm: aut
affiliation:
Högskolan Dalarna
Datateknik
nameIdentifier:
hfl
0000-0002-1429-2345
@attributes:
type: personal
namePart:
Gilani
Syed
role:
roleTerm: aut
@attributes:
type: personal
authority: du
namePart:
Dougherty
Mark
role:
roleTerm: aut
affiliation:
Högskolan Dalarna
Datateknik
nameIdentifier: mdo
originInfo:
dateIssued: 2006
place:
placeTerm: Palma de Mallorca Spain
relatedItem:
@attributes:
type: host
titleInfo:
title: The 10th IASTED International Conference on Artificial Intelligence and Soft Computing
physicalDescription:
form: print
typeOfResource: text