Traffic sign recognition with multi-scale convolutional networks

Pierre Sermanet, Yann LeCun

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We apply Convolutional Networks (ConvNets) to the task of traffic sign classification as part of the GTSRB competition. ConvNets are biologically-inspired multi-stage architectures that automatically learn hierarchies of invariant features. While many popular vision approaches use hand-crafted features such as HOG or SIFT, ConvNets learn features at every level from data that are tuned to the task at hand. The traditional ConvNet architecture was modified by feeding 1nd stage features in addition to 2nd stage features to the classifier. The system yielded the 2nd-best accuracy of 98.97% during phase I of the competition (the best entry obtained 98.98%), above the human performance of 98.81%, using 3232 color input images. Experiments conducted after phase 1 produced a new record of 99.17% by increasing the network capacity, and by using greyscale images instead of color. Interestingly, random features still yielded competitive results (97.33%).

Original languageEnglish (US)
Title of host publication2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program
Pages2809-2813
Number of pages5
DOIs
StatePublished - 2011
Event2011 International Joint Conference on Neural Network, IJCNN 2011 - San Jose, CA, United States
Duration: Jul 31 2011Aug 5 2011

Other

Other2011 International Joint Conference on Neural Network, IJCNN 2011
CountryUnited States
CitySan Jose, CA
Period7/31/118/5/11

Fingerprint

Traffic signs
Color
Classifiers
Experiments

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence

Cite this

Sermanet, P., & LeCun, Y. (2011). Traffic sign recognition with multi-scale convolutional networks. In 2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program (pp. 2809-2813). [6033589] https://doi.org/10.1109/IJCNN.2011.6033589

Traffic sign recognition with multi-scale convolutional networks. / Sermanet, Pierre; LeCun, Yann.

2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program. 2011. p. 2809-2813 6033589.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Sermanet, P & LeCun, Y 2011, Traffic sign recognition with multi-scale convolutional networks. in 2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program., 6033589, pp. 2809-2813, 2011 International Joint Conference on Neural Network, IJCNN 2011, San Jose, CA, United States, 7/31/11. https://doi.org/10.1109/IJCNN.2011.6033589
Sermanet P, LeCun Y. Traffic sign recognition with multi-scale convolutional networks. In 2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program. 2011. p. 2809-2813. 6033589 https://doi.org/10.1109/IJCNN.2011.6033589
Sermanet, Pierre ; LeCun, Yann. / Traffic sign recognition with multi-scale convolutional networks. 2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program. 2011. pp. 2809-2813
@inproceedings{73987af8e88944228e60360906075114,
title = "Traffic sign recognition with multi-scale convolutional networks",
abstract = "We apply Convolutional Networks (ConvNets) to the task of traffic sign classification as part of the GTSRB competition. ConvNets are biologically-inspired multi-stage architectures that automatically learn hierarchies of invariant features. While many popular vision approaches use hand-crafted features such as HOG or SIFT, ConvNets learn features at every level from data that are tuned to the task at hand. The traditional ConvNet architecture was modified by feeding 1nd stage features in addition to 2nd stage features to the classifier. The system yielded the 2nd-best accuracy of 98.97{\%} during phase I of the competition (the best entry obtained 98.98{\%}), above the human performance of 98.81{\%}, using 3232 color input images. Experiments conducted after phase 1 produced a new record of 99.17{\%} by increasing the network capacity, and by using greyscale images instead of color. Interestingly, random features still yielded competitive results (97.33{\%}).",
author = "Pierre Sermanet and Yann LeCun",
year = "2011",
doi = "10.1109/IJCNN.2011.6033589",
language = "English (US)",
isbn = "9781457710865",
pages = "2809--2813",
booktitle = "2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program",

}

TY - GEN

T1 - Traffic sign recognition with multi-scale convolutional networks

AU - Sermanet, Pierre

AU - LeCun, Yann

PY - 2011

Y1 - 2011

N2 - We apply Convolutional Networks (ConvNets) to the task of traffic sign classification as part of the GTSRB competition. ConvNets are biologically-inspired multi-stage architectures that automatically learn hierarchies of invariant features. While many popular vision approaches use hand-crafted features such as HOG or SIFT, ConvNets learn features at every level from data that are tuned to the task at hand. The traditional ConvNet architecture was modified by feeding 1nd stage features in addition to 2nd stage features to the classifier. The system yielded the 2nd-best accuracy of 98.97% during phase I of the competition (the best entry obtained 98.98%), above the human performance of 98.81%, using 3232 color input images. Experiments conducted after phase 1 produced a new record of 99.17% by increasing the network capacity, and by using greyscale images instead of color. Interestingly, random features still yielded competitive results (97.33%).

AB - We apply Convolutional Networks (ConvNets) to the task of traffic sign classification as part of the GTSRB competition. ConvNets are biologically-inspired multi-stage architectures that automatically learn hierarchies of invariant features. While many popular vision approaches use hand-crafted features such as HOG or SIFT, ConvNets learn features at every level from data that are tuned to the task at hand. The traditional ConvNet architecture was modified by feeding 1nd stage features in addition to 2nd stage features to the classifier. The system yielded the 2nd-best accuracy of 98.97% during phase I of the competition (the best entry obtained 98.98%), above the human performance of 98.81%, using 3232 color input images. Experiments conducted after phase 1 produced a new record of 99.17% by increasing the network capacity, and by using greyscale images instead of color. Interestingly, random features still yielded competitive results (97.33%).

UR - http://www.scopus.com/inward/record.url?scp=80054736963&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=80054736963&partnerID=8YFLogxK

U2 - 10.1109/IJCNN.2011.6033589

DO - 10.1109/IJCNN.2011.6033589

M3 - Conference contribution

SN - 9781457710865

SP - 2809

EP - 2813

BT - 2011 International Joint Conference on Neural Networks, IJCNN 2011 - Final Program

ER -