Road Sign Recognition In Real-Time

 Yohay Swirski and Alexander Berkovich
Supervised by Dori Peleg


When driving, recognizing the road signs is crucial for safely reaching one's destination. An automatic system that alerts the driver about road signs may be helpful and might help save lives. In this project, an algorithm which detects and classifies road signs in video movies was created. This algorithm can be modified to work in Real-Time, if connected to a suitable camera and computer.

The Problem

The streets are a jungle. Different Road signs may appear from anywhere in front of the driver. An automatic system that can overcome all these problems and alerts the driver to the presence of these signs can prevent accidents from happening. It might save money and even lives.

The Algorithm

The algorithm follows these classification steps:

The steps of the algorithm are:

  • Color segmentation using HSV color space.
  • Feature extraction.
  • SVM classification of 'red sign', 'blue sign', 'no sign'.
  • Filtering candidates that do not appear at least in 4 out of 5 consecutive frames.
  • Correlation check of the sign with a normalized correlation map.
  • Correlation check of the inner object with a normalized correlation map.


The final statistics of the algorithm, without considering video advantage:

Blue signs:

Red signs:

Classification error - cases in which the algorithm confuses two different signs.

Correlation error - cases in which the correlation of the sign was too low for the algorithm to determine the correct sign.

When considering the fact that a sign can be classified correctly in at least one frame, the final statistics are:

Considering the bad video conditions we faced during the entire project, these results are satisfying.


The algorithm was created in Matlab-7.0 enviorment. In addition, the OSU-SVM toolbox was used.

A video camera was used to capture the movies and create the data base.


An algorithm that detects road signs and alerts to their presence was created, in Matlab environment. If connected to a suitable camera and implemented in fast environmentthis algorithm can work in Real-Time to detect road signs.

The algorithm can overcome problems such as: shadow, different shades of the colors in the sign, partly concealed signs, some disfigure of the sign and appearance from different angles.

In addition, a special effort was made in order to improve the Run-Time of the algorithm. This effort includes some compromises in regard to the system performance

An example of a graphical interface that the algorithm produces:

See demo movie ! (80Mb)


[1] References of signs -

[2] The Interlace problem and possible solutions -

[3] An article about road sign detection -

[4] Bahram Javid, ""Image recognition and classification"", part 4, pages 403-426, 2002

[5] Gal Dayan & Naama Hait- Road Sign Recognition Project Based On SVM Classification -

[6] Visual And Auditory Systems Course

[7] C. Chin-Chung and L. Chin-Jen, LIBSVM: a Library for Support Vector Machines. December, 2002.
LIBSVM Software:

[8] S. R. Gunn, Support Vector Machines for Classification and Regression, Faculty of Engineering and Applied Science Department of Electronics and Computer Science, May 1998.


We would like to thank our supervisor Dori Peleg for his support and guidance throughout this project.

Also we would like to thank Johanan Erez and Ina Krinski that helped us in every technical aspect and hardware issues.

Finally we would like to thank the Ollendorff Minerva Center for supporting the project.