IJSTE - International Journal of Science Technology & Engineering | Volume 3 | Issue 03 | September 2016

ISSN (online): 2349-784X

Traffic Light Recognition System: A Computer

Vision based Approach
Ms. Priyanka Verma
M. Tech. Student
Department of Computer Science & Engineering
Noida Institute of Engineering and Technology, Greater Noida

Mr. Veer Bhadra Pratap Singh

Assistant Professor
Department of Computer Science & Engineering
Noida Institute of Engineering and Technology, Greater Noida

Abstract
This paper states the use of monocular video camera for the detection of traffic lights under different conditions i.e. adverse
weather and low illumination, haze, fog etc. The system proposed in the report incorporates a color pre-processing module to
enhance the discrimination of red and green regions in the image and handle the blooming effect that is often observed in such
scenes. The fast radial symmetry transform is utilized for the detection of traffic light candidates and finally false positive results
are minimized using spatiotemporal persistency verification. The system is qualitatively assessed in various conditions, including
driving in the rain, at night and in city roads with dense traffic, as well as their synergy. It is also quantitatively assessed on a
publicly available manually annotated database, scoring high detection rates.
Keywords: Computer Vision, Traffic Light Detection, ADAS, Monocular Camera
________________________________________________________________________________________________________
I.

INTRODUCTION

Traffic light recognition has become really important for Driver Assistance Systems (DAS). It provides essential information to
the driver on intersections and crosswalks and it can reduce the number of crashes for missed traffic light due to distraction. We
introduce traffic light recognition with status detection concept. The algorithm is evaluated from a generic point of view but the
applications range from Intelligent Transportation System (ITS) to visual impaired and color vision deficiencies aid to safely
cross streets.
A very important part of road rules and traffic management is traffic signal light. It is the main technique and concept which
helps everyone on the road to manoeuvre safely and efficiently. Traffic lights are meant to be followed to be safe and help
everyone on the road to travel coolly. Thus traffic signal detection and recognition has gained importance in the days of today.
People due to negligence tend to cross traffic signals even without getting to know its status. This is a serious issue which causes
a lot of accidents. Also color-blind users not able to recognize common redgreen traffic control signal lights which deprives
them from driving vehicles. Reliable traffic light recognition has been one of the main challenging problems since the
introduction of autonomous vehicles.

Fig. 1: General Traffic Light Signals

In this paper we face the traffic lights (TL) recognition problem: by first reviewing the current systems and public datasets in
the field, and then proposing a system that addresses some of the problems in the other systems. Specifically, we will deal with
suspended and handle supported traffic lights and develop a system able to recognize these ones at daytime. Concerning
detection we find from on-board traffic recognition systems, which aim at automatically detecting the traffic lights in order to
provide useful information to the drive, such as crossroad or crosswalk probability, dangerous area, etc. to the high level decision
taking ADAS.
II. OUTLINES OF THE PROPOSED SYSTEM OVER THE EXISTING SYSTEM
Reliable traffic light recognition has been one of the main challenging problems since the introduction of autonomous vehicles.
Traffic light recognition has become really important for Driver Assistance Systems (DAS). It provides essential information to
the driver on intersections and crosswalks and it can reduce the number of crashes for missed traffic light due to distraction. A

All rights reserved by www.ijste.org

99

Traffic Light Recognition System: A Computer Vision based Approach

(IJSTE/ Volume 3 / Issue 03 / 017)

wearable device could be designed for the visually impaired and the blind to help them safely cross streets. The system proposed
till date is variant of day, haze or bad weather conditions such as fog. The proposed system is invariant of day, night and weather
changes. It is unsupervised and hence generic in its approach. Present study is using single color space and hence lacks correct
detection in low resolution images and videos. Also the present study is not invariant of day and night.
III. PROPOSED METHODOLOGY
The have proposed a robust method for detection of traffic lights. Our algorithm is capable of handling the problem both in day
and night. The algorithm exploits the color spaces for proper detection of traffic lights. The algorithm uses different techniques
to detect traffic lights in day and night. The color of the lights on the camera sensor is different in the day and night. Firstly the
mode of detection is analyzed which will tell us that it is day or night. For this purpose we use the mean sky value. For day time
the value will be high and for the night the value will be low. Once detected we decide which technique to apply for the detection
of the traffic lights.
For the day time we analyze the captured image in the YCbCr and HSV color space. Cr frame is selected and used as base
image for the red color detection. The region of interest is selected for the removal of arears which can not contain the traffic
light. The lower part of the image can never contain the traffic light in the image. The candidates for the traffic light is extracted
and checked for its probable size in pixels. Very big and too small candidates are rejected. For the remaining candidates we find
the YCbCr image and check if mean Cr value of the candidate is more than a threshold (100) and more than its nearby area. If
the candidate passes the test it is taken ahead else it is dropped. For detecting green color in the day we use the HSV color space.
Thresholding the hsv color image gives us the base image for green color detection. Value frame below .4 is suppressed to 0 and
value frame between 0.4 and 0.5 is also suppressed to 0 if the saturation frame is less than 0.4. In the base image region of
interest is extracted. The candidates in this region of interest are checked for their size. The remaining candidates are passed
through a test which checks if the hue and value of the candidate falls in the green region or not. Also as an additional check
mean Cr frame value of the candidate is checked (<120) to validate it. The detected red and green candidates are checked for
symmetry i.e. do the detected ones lie in the same line. The out of line candidates are dropped. This gives us the final traffic
light.
For the night time we us the grayscale frame of the colored image. The image is converted into black and white image and
used as our base image for detection. Further the region of interest is selected. Candidates in the region of interest are tested for
size and proportion. The candidates not applying the defined size and proportions are dropped. For the remaining candidates we
find the mean red, green and blue value inside the candidate and also for the local neighbourhood of the candidate. For detecting
red color we check if red mean inside the candidate is high, the difference between mean red and green color is more (>30) and
the mean red color outside the candidate in the local neighbourhood is less than it. For detecting the green color we check if the
mean green value of the candidate is high, the difference between mean red and green value is more and the mean green color
outside the candidate in the local neighbourhood is less than it. Further the detected candidates are checked for alignment and
proportion. The final detected candidate is the color of the traffic light.
IV. EXPERIMENTAL RESULTS
We have tested our proposed algorithm in various environmental conditions and have got consistent results. The image below is
captured on a cloudy. Here we can see that the tail lights of the car resemble the traffic light but the detection was correct.

Fig. 2(a): Results

Fig. 2(b): Results

The image below was shot in midday. As we can see in the image that there can be many possible false candidates for the red
color traffic light like the red colored car. In this result we have detected all the traffic signs of the road side we are on.

All rights reserved by www.ijste.org

100

Traffic Light Recognition System: A Computer Vision based Approach

(IJSTE/ Volume 3 / Issue 03 / 017)

The image below is shot at night. At night green color can be found near blue lights and thus can create problem but the proper
selection of region of interest and color space helps us detect the light.

Fig. 3(a): Image below is shot at night

Fig. 3(b): Image below is shot at night

The image below is again a night shot where we detect the red traffic light. At night lights sparkle and hence our choice of
different color space helps us detect the light better.
In the image below we show the traffic light sparkle at its extreme.

Fig. 4(a): Image below of the traffic light sparkle

Fig. 5(a): Image below of the traffic light sparkle

Fig. 4(b): image below of the traffic light sparkle

Fig. 5(b): image below of the traffic light sparkle

We have tested our algorithm on more than 10000 frames of video and have got results as in table below
Table 1
Precision and Recall Values
TP FP FN Precision Recall
140 25
8
84.8
94.6

All rights reserved by www.ijste.org

101

Traffic Light Recognition System: A Computer Vision based Approach

(IJSTE/ Volume 3 / Issue 03 / 017)

V. CONCLUSION
Technology for unmanned vehicle is rapidly advanced. Peoples are interested in unmanned vehicle. In urban environment,
unmanned vehicle should pass crossroads. Therefore, signal light detection is important. There are many research and
development for unmanned vehicle. Contest has opened, but, it is artificial and limited. Research for real-time tracking was just
simulation. AI based research has slow run time. HSI color map based research has fast run time. But it has low detection rate. In
this report, we presented the new algorithm for signal light detection. The proposed algorithm has high detection rate. Also, red
and green signal lights are detected successfully. Low price camera can be used for this algorithm. Environment is real urban
road. Condition for weather is not good, but, detectable. And, all of computers and equipments are built in real vehicle.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]

V. Anh, A. Ramanandan, C. Anning, J. A. Farrell, and M. Barth, "Real-Time Computer Vision/DGPS-Aided Inertial Navigation System for Lane-Level
Vehicle Navigation," Intelligent Transportation Systems, IEEE Transactions on, vol. 13, pp. 899-913, 2012.
Z. Cai, M. Gu, and Y. Li, "Real-time arrow traffic light recognition system for intelligent vehicle," Las Vegas, NV, 2012, pp. 848-854.
M. Diaz-Cabrera, P. Cerri, and J. Sanchez-Medina, "Suspended traffic lights detection and distance estimation using color features," Anchorage, AK, 2012,
pp. 1315-1320.
J. Gong, Y. Jiang, G. Xiong, C. Guan, G. Tao, and H. Chen, "The recognition and tracking of traffic lights based on color segmentation and CAMSHIFT
for intelligent vehicles," La Jolla, CA, 2010, pp. 431-435.
M. Omachi and S. Omachi, "Detection of traffic light using structural information," Beijing, 2010, pp. 809-812.
Y. Shen, U. Ozguner, K. Redmill, and J. Liu, "A robust video based traffic light detection algorithm for intelligent vehicles," Xi'an, 2009, pp. 521-526.
H. Tae-Hyun, J. In-Hak, and C. Seong-Ik, "Detection of traffic lights for vision-based car navigation system," vol. 4319 LNCS, ed. Hsinchu, 2006, pp. 682691.
C. Yu, C. Huang, and Y. Lang, "Traffic light detection during day and night conditions by a camera," Beijing, 2010, pp. 821-824.
R. De Charette and F. Nashashibi, "Real time visual traffic lights recognition based on spot light detection and adaptive traffic lights templates," Xi'an,
2009, pp. 358-363.
Y. Li, Z. X. Cai, M. Q. Gu, and Q. Y. Yan, "Traffic lights recognition based on morphology filtering and statistical classification," Shanghai, 2011, pp.
1700-1704.
J. Ying, X. Chen, P. Gao, and Z. Xiong, "A new traffic light detection and recognition algorithm for electronic travel aid," in Intelligent Control and
Information Processing (ICICIP), 2013 Fourth International Conference on, 2013, pp. 644-648
A. Ramananadan, C. Anning, J.A Farrell, V. Anh, M. Barth, Real Time Computer Vision/DGPS-Aided Inertial Navigation System for Lane Level Vehicle
Navigation, Intelligent Transportation Systems, 2012
Mehdi Salarian, Andrea Manavella, Rashid Ansari, A vision based system for Traffic Lights Recognition, 978-1-4673-7606-8/15/$31.00 _c 2015 IEEE.

All rights reserved by www.ijste.org

102