header_image

TRAFFIC AND ROAD SIGNS RECOGNITION

Dr. Hasan Fleyeh
Monday, November 10, 2008
11:00AM - Harris Center 101

Abstract


Traffic sign recognition is one of the important fields in Intelligent Transportation Systems which build a safe, effective and integrated transportation environment based on advanced technologies. A mixture of computer vision and artificial intelligence is used to detect and classify traffic signs in outdoor scenes. The recognition process is achieved by finding the sign in a raw image which is taken by a camera mounted on a moving vehicle, separating it from its surroundings and converting it into a suitable form to enable the system to decide the sign type and its contents. The system is based on combining a sign's colours and shapes in order to recognise the traffic sign. Therefore, it consists of three major parts; a colour segmentation stage, a shape recogniser, and the classifier. Colour segmentation is carried out using a shadow-highlight invariant algorithm using hue as colour information. The algorithm is immune to shadows and/or highlights which can affect the segmentation quality. It is very robust as it was tested using hundreds of images from the raw images library. It works in a very wide range of environmental conditions, and it generates a segmented image which is noise free. The algorithm has the ability to eliminate all small objects retaining the segmented image with the objects to be recognised.

Once the image is segmented, every remaining object is labelled with a unique label using connected components labelling algorithm. These objects are tested using fuzzy shape recogniser which was developed to accept elliptical, triangular, octagonal, and rectangular shapes. Fuzzy rules are used to decide whether a shape belongs to this set of shapes or not. The object is passed to the classifier only when certain conditions are satisfied; the shape belongs to the set of expected shapes, the colour of the signs border passes the test, and the interior of the sign passes the test. An SVM classifier is used for the classification of traffic signs. The classification is carried out in two stages; in the first stage the shape of the sign is classified, and in the second stage the interior of the sign is classified. The classifier is trained and tested using binary images and five different types of moments. The moments used are Geometric moments, Zernike moments, Legendre moments, Orthogonal Fourier-Mellin Moments, and Binary Haar Moments.

Short Bio


Dr. Hasan Fleyeh is vice chair of the computer science department at Dalarna University in Sweden. He received a B.Sc. Electrical Engineering from University of Technology and in 1980 and M.Sc. in Computer Engineering from the same university in 1983. He also received the Ph.D. from Napier University, Scotland, UK in 2008. He worked as a researcher at Astronomy and Space Research Center, Baghdad, Iraq from 1984 to 1990, then became a staff member at Computer Science Department at Baghdad University. He has worked at Dalarna University since 1999. Dr. Fleyeh's field of research is Computer Vision, Image processing and Computer Graphics. His current research interest is in Traffic Sign Recognition. Has has published 15 papers in the last five years and currently supervises several master degree students. He is also arranging a special session in ITS for the 4th Indian International Conference on Artificial Intelligence (IICAI-09).

FEEDBACK | Webmaster | EECS | FSI | CECS | UCF
University Of Central Florida | Orlando, Florida 32816-2362 Phone: 407-823-2341