images of traffic scenes and road signs
Following images show the traffic scenes grabbed from the moving car.
These images are the input of the road sign recognition system. The detection algorithm
searches for geometrical shapes there by the use of local orientations.
Let us discuss another traffic scene image. It contains two road signs. These are
scaled two times (four times after click) on the next picture. Notice the interlace in the
To learn the statistical classifier to label real road sign images properly, one
needs to use large databases of sign images. The reason is the statistical nature of
used machine learning algorithms - the more information at hand, the better
generalization capabilities of the classifier.
To collect such a base, we have used
Olympus Camedia digital device. Following images show source images that
are manually cropped and put into our pattern database. The last image in the line
contains several images from the database - various sign types in many sizes are present.
These images are used for the learning of our road sign classifier.
Last modified: Wed Feb 7 16:38:40 CET 1996