Abstract:
Automated traffic sign recognition is an important part of a driver assistance system for
future automobiles. In this work, algorithms were developed in a MATLAB environment
to recognize traffic signs from recorded video sequences while the vehicle is in motion.
The work focused mainly on the development of a fast algorithm to recognize traffic
warning signs placed on local roads.
In general, the detection of an object in a video scene in a natural environment has two
major obstacles: (1) Locating the desired object when the traffic scene is composed of
many objects such as vehicles, buildings, pedestrians, name boards, etc., and (2) Varying
illumination due to changing weather conditions, brightness conditions and shadows. The
method adopted in this work detects the location of the sign in a single image frame based
on its color information. Samsung digital camera having standard definition video
recording (640x480) at 30 frames per second was used to collect short video clips required
for this work. In order to avoid the effects due to brightness variation, color space
conversion, from RGB to normalized RGB, is used. Morphological operations were used
to remove the noise from image frames to improve the sign extraction process. To classify
the warning signs, a simple technique based on the geometrical characteristics of the signs
were adopted.
The results of this work showed high performance in the extraction and recognition of
traffic signs on local roads. The algorithm was tested on a variety of video segments and
showed 98% accuracy in extracting and recognizing selected traffic warning signs during
the daytime
