Professional Documents
Culture Documents
NAYANA. T. N
Student, M.tech (Digital Signal Processing)
Centre for Emerging Technologies, Jain University, Kanakapura, Karnataka State, India
MR. DEEPAK.V.INGALE
Assistant Professor, Centre for Emerging Technologies,
Jain University, Kanakapura, Karnataka State, India
ABSTRACT: According to modern requirement for vehicle safety, it must be noted that effective
driving is often dependent upon computer systems the vehicles are equipped with, rather than the
driver. Therefore, the main measure to reduce road accidents is to be taken in the development of
the systems that warns the driver of the road conditions, well ahead of time. This can be achieved,
by developing an automated system that would detect traffic signs along the way in advance, and
inform the driver about driving conditions, without distracting him in his driving. This paper
proposes a system for detection of traffic signs from still images or real time video. Algorithm is
developed by using Hue, Saturation and Intensity (HSI) color space, to segment the road signs and
Traffic signs detected by using standard deviation of segmented region as threshold value.
Experimental results on the accuracy of the road sign detection are reported in this paper. The
proposed algorithm is simulated using MATLAB.
KEYWORDS: Traffic Signs, Traffic sign detection, Segmentation, HSI color space, Median
Filter.
INTRODUCTION
Intelligent vehicles are becoming essential part of our day to day life. A large number of accidents
on roads occur due to the carelessness of drivers and violation of the traffic rules. The function of
traffic signs can be broadly classified as: i. to control the traffic, and ii. To warn about the status of
the road ahead. The main measure to reduce road accidents is to be taken in the development of the
systems that warns the driver of the road conditions, well ahead of time. This can be achieved, by
developing an automated system that would detect traffic signs along the way in advance, and
inform the driver about driving conditions, without distracting him in his driving.
RELATED WORK
Considerable volume of research has been carried out, and is ongoing, for designing a robust
traffic sign recognition system.
Bascon et. al [1] detected the road signs using color segmentation. Traffic signs are classified by
shape using linear SVMs and content recognition based on Gaussian Kernel SVMs.
Martinovic et. al [2] described Viola Jones detection algorithm for detecting the road signs. The
detected road signs are classified using Neural Networks.
Ming Liang et. al [3] proposed a procedure with two modules: Region-of-Interest (ROI) extraction
module and Recognition Module. ROI module detects the traffic signs and recognition module
extracts histogram descriptors from the ROIs to provide a reliable identification of traffic sign and
uses SVMs to judge whether an ROI is a target sign or not.
Whelan et.al [4] proposed a system which consists of three main blocks: Color Segmentation,
Candidate Traffic Signs Validation and Traffic Sign Classification. Candidate regions obtained by
color segmentation are validated with respect to strict shape criteria to generate scale invariant
descriptors and are classified using SVM classifiers.
Yusoff et. al [5] described the road sign detection method by Color Segmentation and traffic signs
are classified using Shape Classification, Symbol Recognition based on extent value, area and
perimeter ratios with the template values.
Gauri et. al [6] proposed a system where color thresholding technique is used to produce binary
image , blob detection method is used to detect the traffic signs and are classified using template
matching.
Yixin Chen et. al [7] described the method which consists of HSV color space based Image
Segmentation, geometry features based detection and Affine transformation correction, extraction
of Eigen vectors using Gabor filter, Classification and recognition by SVM.
METHODOLOGY
Block diagram of the proposed system for the detection of traffic sign is shown in Figure 1. It
mainly consists of two stages: i. RGB to HSV Conversion, and ii. Sign detection using Image
Segmentation.
The captured images are converted to HSV color space. The HSV color space allows decoupling
the color, saturation and intensity information [6].
Image Enhancement is achieved using Median filter and image is segmented to generate the binary
image. Morphological operation is performed to close any open area. Region properties are applied
to get boundary box properties of the ROI of the candidate object.
RGB color space is the combination of 3 primary colors: Red, Green and Blue, commonly used
by most computer devices. Color space that corresponds more naturally to human perception is the
HSV color spaces [6]. The Hue (H) element represents the perceived colors (red, yellow and blue
for road signs), which is within the interval [0, 360]. The Saturation(S) represents how much of
white light has been added to a pure color, and is within the interval [0, 255] [5]. HSV space is
136
more suitable for image segmentation, because H and S components in general are not affected by
light intensity [7]. Hence HSV space is chosen in the proposed scheme. The RGB color spaces are
converted to HSI color spaces by using the following equations [7].
SIGN IMAGE
DETECTION SEGMENTATION
M= max(R,G,B)
m= min(R,G,B)
C=M-m
undefined if C 0
G B
mod 6 if M R
C
H' B-R
C 2 if M G
R-G 4 if M B
C
H 60 H '
The saturation component is given by the following: [7]
0 if M 0
SC m
M 1 M otherwise
137
IMAGE SEGMENTATION
Before segmentation, image enhancement is performed using Median filter, as it removes the noise
without blurring of edges. Median filters are non linear filters and give better results than
averaging filters. Here, it is assumed that the hue and saturation components of the HSV space are
sufficient to isolate traffic sign from the image. Hence S component of HSV space is converted to
a binary image by using simple Otsus method, where, the intra-class variance of the black and
white pixels is minimized by choosing a suitable threshold value. The segmented (binary) image is
shown in figure 2 (b). As shown in this figure, the triangular red rim of the traffic sign has been
correctly identified, but at the same time other objects such as the cladding of the buildings, small
sections from an advertising panel are also highlighted in the output image. Thus, morphological
operation is performed to minimize these background noises, as shown in figure 2 (c).
(a) (b)
(c) (d)
Further, for the resulting image, selective region properties are applied to get bounding box of the
ROI of the candidate region. This results in the detection of traffic sign as shown in figure 2 (d).
EXPERIMENTAL RESULTS
In order to evaluate the proposed system, numerous traffic signs were captured. The captured
images were processed using the proposed algorithm and results are shown below. MATLAB tool
is used to design and implement the traffic sign detection. Results are shown in figure 3.
138
(a)
(b)
(c)
Traffic sign detection based on segmentation is presented in this paper. Firstly, RGB images are
converted to HSV color space and traffic signs based on S component threshold are detected.
The results are satisfactory at this stage.
Some improvements listed below, can enhance the performance further and can be recommended
for future work.
An efficient algorithm for classification of detected traffic signs could be implemented.
The algorithm may be further tested under extreme weather conditions, and any necessary
improvements can be implemented.
Suitable hardware may be developed using suitable micro-controller platform for Real-time
implementation.
139
REFERENCES
140