Digital Signal Processing Reference
In-Depth Information
Investigation of Directional Traffic Sign Feature
Extracting Based on PCNN in Different Color Space
*
Mengjun Wang
**
, Lu Yang, Xia Wang, and Jianfei Liu
School of Information Engineering,
Hebei University of Technology, Tianjin, 300401, China
{Mengjun Wang,wangmengjun}@hebut.edu.cn
Abstract.
For the first time in this paper, directional traffic signs feature ex-
tracting based on Pulse Coupled Neural Network (PCNN) in different color
space are investigated. Entropy series is extracted from the image of traffic sign
in both RGB model and HSV model. Each entropy series of
R, G, B, H, S, V
color space is used as feature vector for recognition, match analysis is carried
out by minimum variance. Experiments are carried out based on the directional
signs class in national standard GB5768-1999 database. Experiment results
show that feature vector based on Entropy series in
B
color space get the higher
recognition rates than the other color space, with 50 iteration and 5 × 5 convolu-
tion kernel matrix of PCNN.
Keywords:
Traffic sign recognition, Pulse Coupled Neural Network, entropy
sequence, color space.
1
Introduction
Traffic sign recognition (TSR) is an important part of the Intelligent Transportation
Systems (ITS), the recognition result is not only to remind drivers of traffic safety in-
formation, but also be used as the autopilot overall coordination of the joint input. For
traffic sign recognition system, feature extraction is the identification process. The
most obvious features are color and shape information. For example, color information
is used at HSV model and RGB model [1-3]. But color information is only used in the
detection step. Shape information is used to detect the round, triangle and hexagon for
traffic signs [3-5].
PCNN is derived from the Echorn's neuron model. Echorn's neuron model is devel-
oped by simulating the activities of the visual nerve cell based on the observation of
the visual cortex nerve cell of cats. Researches on PCNN and its applications have
developed greatly in recent years, it can used to image segmentation, noise reduction,
image smoothness, and feature extraction. Pulse coupled neural network (PCNN) has
been used to deal with the traffic signs [6, 7]. While extracting the features, gray im-
ages is adopted. Color information is more important in TSR system because of the
*
Project supported by College Technology and Research Youth Foundation of Hebei Province
(No. 2010121).
**
Corresponding author.
