You are on page 1of 4

International Conference on Emerging Research in Electronics, Computer Science and Technology - 2015

Traffic Signal Time Analysis and Voice - Based App


for Visually Impaired Pedestrians
Basavaraju R Dr. Chetana Hegde
Research Scholar - VTU RRC, Associate Professor, Department ofMCA,
Visvesvraya Technological University, RNS Institute of Technology,
Bangalore, India Bangalore, India
basavaraj u.revanna@gmail.com chetanahegde@ ieee.org

Abstract- Analysis of traffic signal and ensure the safety of pedestrian analyzes the street which he/she need to head
visually impaired people is a major challenge in helping disabled towards, he/she cannot proceed further when there is a traffic
people. In this paper, we propose development of an application signal. It is necessary for such a people to understand when
software which can be easily installed on a mobile device there is a green signal to cross the road.
equipped with a camera. This application opens a camera and
captures the traffic timer display upon tapping the app-icon. The Given a scenario where traffic signals are possessing timer
timer image is then processed to segment the digits in them to display, it would be great if the blind pedestrian is informed
identify the actual time in the numeric form. The detected time is how much time is remaining for green signal to tum-on. If the
then converted as a voice message and played using the app. blind pedestrian is informed the time remaining through voice,
Thus, the visually impaired person can hear the message about he/she can wait and then safely cross the road.
the time left to turn on the pedestrian signal. Getting this alert Keeping the problems faced by visually impaired
message, he/she can safely cross the road. The accuracy of the pedestrians in mind, this paper proposes development of an
proposed algorithm is found to be 100% as it detected all the Android based application involving image processing
digits in every instance of the timer display image.
techniques. A novel algorithm is proposed to capture the timer
Index Terms- Android SDK, connected components, display as an image, process it to recognize the actual time and
correlation coefficient, histogram, preprocessing, segmentation, to inform the time to the user through voice. The paper is
text to speech conversion. organized as follows : Section II deals with the related work and
Section III presents the architecture and model. Section IV
describes the implementation of the proposed algorithm and the
I. INTRODUCTION
perfonnance analysis. Section V concludes the work.
It is a well known fact all over the world that the visually
impaired (partially or completely blind) people face difficulty II . RELATED WORK
in crossing the roads. Visually impaired people can travel or A brief survey of the related work in the area of various
cross the street with the help of other pedestrians, using a white aids available for blind pedestrians as well as the analysis of
cane to identity and avoid obstacles, using a dog guide, using traffic signal is presented in this section.
special optical or electronic aids etc [1]. Crossing the In 1980's, the scientists at Nottingham University proposed
road/traffic signal for visually impaired people involves the idea of having a small device containing plastic/metal cone
following tasks [1]: to be attached at the pole on the pedestrian crossing [2]. When
Locating the Streets: Understanding when they reach a blind person needs to cross the road, he/she has to go near to
the street. This may include a combination of cues, that pole and press the cone. Now, the cone starts spinning and
curb, slopes, traffic sound etc. the signal is made to the vehicles to stop. The spinning of a
Street Recognition: Blind people has to recognize cone will ensure the pedestrian to cross the read safely. But,
which street they have reached to using their mental proposal has many disadvantages: (i) The blind pedestrian has
map; may be by counting the steps, number of turns to locate the pole having this cone to use it. (ii) It may not be
taken etc. possible to set up this device at all the signals. (iii) Improper
Intersection Assessment: Obtaining possible functionality of the device, in case of damage, will put the
intersection of roads, number of streets present in the blind pedestrian in a dangerous situation.
intersection, which exact street they need to travel etc. These push button cones have then been enhanced to
Crossing the Road: After getting the above generate the locator tone [1, 3]. The locator tone infonns
infonnation, facing towards proper street and then pedestrians of the need to push a button and provides an
crossing the road. audible cue to the location of the pushbutton. During the street
crossing the locator tone may be audible in the last lane and
However, these tasks are not so easy for blind pedestrians, provide a cue to the direction of the destination. But, to use this
especially, when there are traffic signals. Even if a blind device also, one has to go near to the device mounted on the

978-1-4673-9563-2/15/$31.00 2015 IEEE

477
International Conference on Emerging Research in Electronics, Computer Science and Technology - 2015

pole. Moreover, it will be difficult to hear the tone in busy the image of a timer display at the traffic. Since, the blind
traffic with lot of vehicles moving around. Also, if there are people will be using the device, there is no guarantee that the
intersection of roads, it may be confusing to the pedestrian that image actually contains the time. Hence, through the pattern
to which road, the tone is beeping the alert. matching algorithm, it is ensured that the image contains the
Researchers at Kyoto Institute of Technology in Japan have timer display. Also, by enhancing techniques, region of interest
developed an electronic eye system to help the blind people to (ROI) is cropped from the original image. For simulation
cross the road at busy intersections [4]. The system consists of purpose, images of ordinary digital clock are considered. The
a camera and a small computer that can detect pedestrian image after extracting ROI is as shown in Figure 2 (a).
crossings, including the width of each crossing and the status
of its traffic lights. Instructions are relayed to the user by
means of a voice speech system with a built-in speaker. But,
this requires the blind pedestrian to carry this system with
him/her may in the form of wearable device, which may not be
affordable by every individual. Moreover, this system has been
tested on 196 intersections, out of which 194 have been (a) ROI extracted (b) Seperator Removed
detected correctly yielding an accuracy of98.97%. Figure 2. Image of digital clock display
Several efforts have been made by the researchers [5 - 10]
B. Preprocessing
to monitor the traffic signal by analyzing green/red lights
through image processing techniques and neural networks. But, Preprocessing involves series of steps. Initially, the RGB
none of these researches have focused to aid visually impaired image has to be converted into binary image for reducing the
people. But, the study [11] about number of road accidents future complexity in the computation. It is a known fact that
faced by visually impaired people enforces reliable research to the digital clock images or timer displays have hour/minute
be carried out to aid them. seperator in the form of colon symbol. For identitying digits in
The proposed method is found to be 100% accurate in the image, we need to remove these seperators. It involves
detecting the time left-out in the signal and hence aiding the following steps -
visually impaired people to cross the road. The proposed (i) Determining the Connected Components: Initially,
method is discussed in the following sections. connected components in a binary image are extracted. Let Y
represent a connected component contained in a set A and
III. ARCHETECTURE AND MODELLING assume that a point p of Y is known. Then the equation (1)
Initially, a database has to be setup containing 10 images containing interative expression yields all the elements of Y -
depicting every digit from 0 to 9. Given an image containing Xk = (X k _ 1 EBB)nA, k=J, 2, 3, . (1)
timer display (ex. Digital clock), obtaining the time involves Where Xo = P and B is a suitable structuring element. If X k =
identitying each digit as a major module. Then all the digits X k j , then the algorithm converges and we get Y = X k
have to be combined to make a sequence to represent the time (ii) Compute area of each component: Area of each
as a text, which, then needs to be converted into voice. The component is computed by counting number of pixels in the
architectural diagram for the entire process is as shown in regaion.
Figure 1. The various steps involved are explained hereunder. (iii) Remove small objects: Small objects in the set A
containing many connected components are removed if the
area of any component is less than a threshold value.
Digital Clock The above three steps remove the seperator as shown in
lmage
Acquisition Figure 2(b). Later, morphological closing is applied to fuse the
narrow breaks and long thin gulfs, to eleminate small holes and
to fill the gaps in the contour in the image. This operation will
ensure that each digit is one connected component. Closing is
Database
done using the equation -
A. B = (A EB B)eB (2)
Here, A is the image and B is the structuring element. The
image after closing operation is as shown in Figure 3.

Figure I. Architectural Diagram

A. Jmage Acquisition and ROJ Detection


The Android based application developed for the proposed
algorithm opens the camera of mobile phone when tapped on
the app-icon. The person holding the mobile device can capture
03 18 Figure 3. Image after closing

478
International Conference on Emerging Research in Electronics, Computer Science and Technology - 2015

C. Digit Segmentation
To segment each digit from the whole image, initially,
connected components from the preprocessed image are (5)
extracted. Then, each digit is segregated by iteratively hiding
the connected components and using the bouding box of each
row in the respective region. This procedure is carried out 1=1 1=1
using following steps - Here,
(i) Search for the next unlabeled pixel, p .
(ii) Use a flood-fill algorithm to label all the pixels in the
connected component containing p.
(iii) Repeat steps (i) and (ii) until all the pixels are labelled.
6ooor-----,-----~------~----_r----~----~

The images containing individual digits are saved with


approriate file names. Digits extracted from the preprocessed 5000
image are shown in Figure 4(a) - (d).
4000

3000

2000

1000

Figure 4. Images containing extracted digits 0~--750~--~1700~--~15~0~--~


20~0---~2~5~
0 ----~300
Range of Intensity va lue
D. Pattern Matching
Figure 5. Histogram obtained for the the digit One
This steps involves comparing extracted digits in the
previous step with the stored images in the database. 6000.------r----~------~----~----~~--__,
Histograms for each of the extracted digits is plotted. For
intensity images, the n bins of the histogram are each half-open
5000
intervals of width A/(n-l) . That is, for intensity images having
float values, the / ' bin is the half-open interval given by -
4000
A(p -1.5) A(p - 0.5)
----''''--------'- ::; X ::; -=----'- (3)
n-l n-l 3000
where x is the intensity value. For intensity images having
integer values, the i h bin is the half-open interval given by - 2000

A(p -1.5) _ 32768::; X::; A(p - 0.5) _ 32768 (4) 1000


n-l n-l
where x is the intensity value. The scalar factor A depends on ..II
0~---~50~--~1~0~0----~1~50~--~
20~0~-~25~0----~
300
the data type of image class and it may be one among the
Range of Inten sity valu e
values 1,255 or 65535. Histograms plotted for two digits 1 and
8 are shown in Figure 5 and Figure 6. Figure 6. Histogram obtained for the the digit Eight
Histogram computed for each extracted digit is compared
with every digit image in the database. The correlation E. Time Detection and Text to Voice Conversion
coefficient is computed between every pair of histograms. If HI
The detected digits can be stored as numbers and hence the
and H2 are histrograms of two images, then the correlation
time can be recognized and displayed. The result in numeric
coefficient between them is calculated using the Equation (5).
format is displayed as -
If the correlation coefficient r is nearer to 1, then we can
The time remaining: 3 minutes 18 seconds
say that the two digits are the same. Hence, we can identify the
But, to aid the blind people, the time has to be informed
actual digit stored in the extracted image.
through voice. Hence, the time in number format has to be
stored as a vector of strings. This text message is then
converted to speech.

479
International Conference on Emerging Research in Electronics, Computer Science and Technology - 2015

IV. IMPLEMENTA nON AND PERFORMANCE ANALYSIS REFERENCES

The proposed technique is implemented using different [1] Accessible Pedestrian Signals - A Guide to Best Practices:
technologies. The initial steps of image capturing and ROI http://www.apsguide.org/index.cfin
detection are done using Java language in Android SDK. Later, [2] The Secret Button at Pedestrian Crossings - A series on small
image processing techniques like preprocessing, digit but significant insights into disabled life:
seperation, pattern matching using histogram and correlation http://www.bbc.comlnews/blogs-ouch-22706881
coefficient are done using MatLab 7.5. Again, for text to voice [3] MUTCD (Manual on Uniform Traffic Control Devices),
conversion, the built-in class TextToSpeech within Android Developed by U S Department of Transportation in 2009:
SDK is used. http://mutcd.thwa.dot.gov/pdfs12009/mutcd2009edition.pdf
The proposed algorithm is tested on 200 digital clock [4] Helping the Blind Cross the Road :
images as well as real-time traffic timer displays. The images http://www.geek.comlscience/helping-the-blind-cross-the-road-
5569941
were acquired with different orientation of the camera. Up to
150 degrees of orientation (angle between camera and the timer [5] D Rosenbaum, J Leitloff, F Kur z, 0 Meynberg and T Reize,
"Real - time Image Processing for Road Traffic Data
display), the digits were identified correctly. Beyond 150, the
Extraction from Aerial Images ", Proc. of ISPRS TC VII
image can not be captured properly and hence the efficiency Symposium, IAPRS, vol. XXXVIII, Part 7B, pp. 469 - 474,
reduces. It is obvious that, in real time situation, the camera 2010.
orientation do not cross 150, hence the proposed algorithm [6] P Gupta, G NPurohit, SPandey, "Traffic Load Computation for
gives lOO% accuracy in detecting the time. Real Time Traffic Signal Control ", Int. Journal of Engineering
and Advanced Technology (HEAT), vol. 2, no. 4, pp. 891 - 895,
V. CONCLUSIONS 2013 .
In this paper, we propose an efficient way of extracting [7] M Puthran, S Puthur, R Dharulkar, "Smart Traffic Signal ", Int.
exact time left for a traffic signal to tum green/red. The Journal of Comp Sc and Info Tech, vol. 6, no. 2, pp. 1360 -
algorithm focuses on the real-time problem faced by visually 1363, 2015.
impaired people in crossing the traffic signals. Hence, a mobile [8] A Salhi, B Minaout, M Fakir, "Robust Automatic Traffic Signs
phone installed with the proposed application will act as a Recognition using Fast Polygonal Approximation of Digital
reliable aid to such people. The proposed method is easily Curves and Neural Networks ", Int. Journal of Adv Comp Sc and
adoptable as it addresses real-life challenges of visually Applications, Special Issue on Advances in Vehicular Ad hoc
impaired people. Networking and Applications, 2014.
http://dx.doi.org/10.14569/Speciallssue.2014.
Though the proposed technique is partially developed using
Android SDK, it can be even implemented using other [9] A. Lorsakul and J. Suthakorn, "Traffic Sign Recognition for
Intelligent Vehicle/Driver Assistance System Using Neural
Operating Systems as well. Also, the main focus in the
Network on OpenCV", Proc. of 4th International Conference on
proposed algorithm is to capture the image containing digits Ubiquitous Robots and Ambient Intelligence (URAl 2007).
and to identify them. Hence, it can be used in various other
[10] V Lepetit, P Fua, "Keypoint Recognition using Randomized
applications also. A video capturing the time in stop-clock can Trees ", IEEE Trans on Pattern Analysis and Machine
be processed to get a series of timer count to tum a light on/off Intelligence, vol. 28, no. 9,pp. 1465 - 1479.
at specific time, for automation of home-appliances and such [11] B Matthews, D Hibberd and 0 Cartsen, "Road and Street
other scientific applications. Crossings for blind and partially sighted people: The
Importance of being certain ", University of Leeds.
http://www.guidedogs.org.uklmedial7187843/road-and-street-
crossings-the-importance-of-being-certain2-2.pdf

480

You might also like