International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072

Smart Traffic Control System using YOLO

Pranav Shinde1, Srinandini Yadav2, Shivani Rudrake3, Pravin Kumbhar4
1,2,3,4Bachelor of Computer, Dept. of Computer Engineering, BSIOTR College, Maharashtra, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Traffic congestion is becoming a serious problem crashes by 8% to 50%. Carnis and Blais showed in their
with large number of cars in the roads. Vehicles queue length research that the initial response to France’s implementation
waiting to be processed at the intersection is rising sharply of their automated speed enforcement program (ASEP) was
with the increase of the traffic flow, and the traditional traffic a 21% reduction in fatal car accidents and a 26% reduction
lights cannot efficiently schedule it. A real time traffic light in non-fatal car accidents.
control algorithm based on the traffic flow is proposed in this
paper. In fact, we use computer vision and machine learning to 1.1 Yolo Real-time Object Tracking in Video
have the characteristics of the competing traffic flows at the
signalized road intersection. This is done by a state-of-the-art, YOLO is a fast, accurate object detector, in First, an
real-time object detection based on a deep Convolutional image is taken and YOLO algorithm is applied. In our
Neural Networks called You Only Look Once (YOLO). Then example, the image is divided as grids of 3x3 matrixes. We
traffic signal phases are optimized according to collected data, can divide the image into any number grids, depending on
mainly queue density and waiting time per vehicle, to enable the complexity of the image. Once the image is divided, each
as much as more vehicles to pass safely with minimum waiting grid undergoes classification and localization of the object.
time. YOLO can be implemented on embedded controller using The objectness or the confidence score of each grid is found.
Transfer Learning technique, which makes it possible to If there is no proper object found in the grid, then the
perform Deep Neural Network on limited hardware resources. objectness and bounding box value of the grid will be zero or
if there found an object in the grid then the objectness will
Key Words: Transfer Learning Technique, YOLO, Deep be 1 and the bounding box value will be its corresponding
Neural Network, Computer vision, Convolutional Neural bounding values of the found object. The bounding box
network. prediction is explained as follows. Also, Anchor boxes are
used to increase the accuracy of object detection which also
1. INTRODUCTION explained below in detail.

Traffic congestion is a major problem in many cities, and 2. PROPOSED SYSTEM ARCHITECTURE:
the fixed-cycle light signal controllers are not resolving the
high waiting time in the intersection. We see often a
policeman managing the movements instead of the traffic
light. He see roads status and decides the allowed duration of
each direction. This human achievement encourages us to
create a smart Traffic light control taking into account the
real time traffic condition and smartly manage the
intersection. To implement such a system, we need two main
parts: eyes to watch the real-time road condition and a brain
to process it. A traffic signal system at its core has two major
tasks: move as many users through the intersection as
possible doing this with as little conflict between these users
as possible.

Video is a powerful medium for conveying

information and data is plentiful wherever there are
cameras. From dash, body, and traffic cams, to YouTube and
other social media sites, there is no shortage of video data.
Interesting applications that exploit this are ripe for
development. One particularly compelling domain where
video analytics has tremendous potential is in automated
traffic control and public safety systems. The mere presence Fig -1: Architecture Diagram
of automated traffic control systems, like red-light and speed
cameras, has been shown to have a positive effect on the
reduction of traffic violations. A worldwide analysis of 28
studies has concluded that such systems have reduced

© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 967
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072

The above figure signify the working of module using cloud, approach. YOLO is a clever convolutional neural network
sever and embedded system. The background detector takes (CNN) for doing object detection in real-time. The algorithm
help of YOLO framework for object detection. In our system applies a single neural network to the full image, and then
we have manipulated YOLO for only detecting vechiles of divides the image into regions and predicts bounding boxes
different categories i.e Truck Car Bike etc. The Yolo is and probabilities for each region. These bounding boxes are
deployed on sever which is connected to embedded system weighted by the predicted probabilities. YOLO is popular
through AWS cloud. The Raspberry pi board is attached to because it achieves high accuracy while also being able to run
onsite camera and the relay attached to signals. As we need in real-time. The algorithm “only looks once” at the image in
internet to connect raspberry pi with cloud we will have to the sense that it requires only one forward propagation pass
connect it using wifi module attached to it. Raspberry pi through the neural network to make predictions.
helps to transfer video from camera to cloud. The
background detector comprises of algorithm with YOLO Our network has 24 convolutional layers followed by 2 fully
module to make out lane density. It is smart way with great connected layers. Instead of the inception modules used by
state of art YOLO framework to manage the control flow of GoogLeNet, we simply use 1 × 1 reduction layers followed by
traffic. 3 × 3 convolutional layers .Fast YOLO uses a neural network
with fewer convolutional layers (9 instead of 24) and fewer
2.1 Hardware Module: filters in those layers. Other than the size of the network, all
training and testing parameters are the same between YOLO
a) Raspberry Pi and Fast YOLO.

b) Sony IMX219 sensor 3.2 DNN :

2.2. Software Module: A deep neural network (DNN) is an artificial neural
network (ANN) with multiple layers between the input and
a) Flask API. output layers. The DNN finds the correct mathematical
manipulation to turn the input into the output, whether it be
b) AWS Cloud.
a linear relationship or a non-linear relationship. The
c) Anaconda network moves through the layers calculating the
probability of each output. For example, a DNN that is
d) REST API trained to recognize dog breeds will go over the given image
and calculate the probability that the dog in the image is a
3. PROPOSED SYSTEM TECHNIQUE’s : certain breed. The user can review the results and select
which probabilities the network should display (above a
3.1 YOLO: certain threshold, etc.) and return the proposed label. Each
mathematical manipulation as such is considered a layer,
and complex DNN have many layers, hence the name "deep"
networks.

DNNs can model complex non-linear relationships.

DNN architectures generate compositional models where the
object is expressed as a layered composition
of primitives.[106] The extra layers enable composition of
features from lower layers, potentially modeling complex
data with fewer units than a similarly performing shallow
network.

DNNs are typically feedforward networks in which

data flows from the input layer to the output layer without
looping back. At first, the DNN creates a map of virtual
neurons and assigns random numerical values, or "weights",
Object detection is one of the classical problems in computer to connections between them. The weights and inputs are
vision where you work to recognize what and where — multiplied and return an output between 0 and 1. If the
specifically what objects are inside a given image and also network did not accurately recognize a particular pattern, an
where they are in the image. The problem of object detection algorithm would adjust the weights.[107] That way the
is more complex than classification, which also can recognize algorithm can make certain parameters more influential,
objects but doesn’t indicate where the object is located in the until it determines the correct mathematical manipulation to
image. In addition, classification doesn’t work on images fully process the data.
containing more than one object. YOLO uses a totally different

© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 968
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072

3.3 CNN:-
A convolutional neural network consists of an input
and an output layer, as well as multiple hidden layers. The
hidden layers of a CNN typically consist of a series of
convolutional layers that convolve with a multiplication or
other dot product. The activation function is commonly
a RELU layer, and is subsequently followed by additional
convolutions such as pooling layers, fully connected layers
and normalization layers, referred to as hidden layers
because their inputs and outputs are masked by the
activation function and final convolution. The final
convolution, in turn, often involves backpropagation in order
to more accurately weight the end product.

Though the layers are colloquially referred to as

convolutions, this is only by convention. Mathematically, it is
technically a sliding dot product or cross-correlation. This
has significance for the indices in the matrix, in that it affects
how weight is determined at a specific index point.

4. Proposed System Control Flow:-

Initially the video will be streaming on all four roads
of the traffic circle. The values will be read frame by frame in
the streaming video of these roads. Camera sends all the
captured input videos to the processing engine on the cloud.
Once the video are received by cloud the background
algorithm filters the video to count the density of vehicles in
every lane. Then based on threshold values each lane is
prioritized and hence timer is set appropriately. Yolo helps
to count vehicle in real time with the speed of 24fps and
combining it with background algorithm .Density is counted
on the basis of vehicle category for transport vehicle and
containers will have more density as compared to vehicle.
Here the total density is counted for every lane. There is a
threshold value based on two criteria one is the density
count of the lane and other is priority of that lane.

Consider lane L1, L2, L3, L4 every lane will have

the equal time Tmax which signify till what maximum time
green signal of particular lane will be lit. The signals will turn
green lane by lane so that there wont be longer waiting time Fig:- Proposed System Control Flow
for any lane with less density. If there is very less density
means if the lane is empty then no green signal will be 3. CONCLUSIONS
turned on for that lane till its density count reaches the
threshold. The video of the lanes which have red signal will The goal of this work is to improve intelligent transport
be recorded and sent to cloud for processing, first the video systems by developing a Self-adaptive algorithm to control
is formatted according to resolution of S x S size. Then the road traffic based on deep Learning. This new system
yolo with the background detector performs Object facilitates the movement of cars in intersections, resulting in
detection on it then density count is calculated and priorities reducing congestion, less CO2 emissions, etc. The richness
of lane are fixed. In this way it will be efficient for automated that video data provides highlights the importance of
traffic control using raspberry pi with Deep learning. advancing the state-of-the-art in object detection,
classification and tracking for real-time applications. There
has been a steady progression of image detection techniques
beginning with feature descriptors like HOG and, more
recently, deep network-based approaches like Faster R-CNN
and YOLO. YOLO provides extremely fast inference speed

© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 969
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072

with slight compromise in accuracy, especially at lower

resolutions and with smaller objects. While real-time
inference is possible, applications that utilize edge devices
still require improvements in either the architecture’s design
or edge device’s hardware.

Finally, we have proposed new algorithm taking this

real-time data from YOLO and optimizing phases in order to
reduce cars waiting time.

ACKNOWLEDGEMENT
The authors would like to thank Mrs. Sanchika Bajpai and
Mrs. Gauri Virkar for their valuable contribution to this
project.

REFERENCES
1. G. Dimitrakopoulos and P. Demestichas, “Intelligent
transportation systems,” IEEE Vehicular Technology
Magazine, vol. 5, no. 1, pp. 77–84, 2010.

2. K. Morinaga, “Car navigation system,” May 16 1995,

uS Patent 5,416,478.

3. K. H. Molloy, J. P. Ward, and V. M. Benson, “Traffic

signal control system,” Jul. 4 1972, uS Patent
3,675,196.

4. R. Lotufo, A. Morgan, and A. Johnson, “Automatic

number-plate recognition,” in Image Analysis for
Transport Applications, IEE Colloquium on. IET,
1990, pp. 6–1.

5. M. Cremer and M. Papageorgiou, “Parameter

identification for a traffic flow model,” Automatica,
vol. 17, no. 6, pp. 837–843, 1981.

6. J. Gajda, R. Sroka, M. Stencel, A. Wajda, and T. Zeglen,

“A vehicle classification based on inductive loop
detectors,” in Instrumentation and Measurement
Technology Conference, 2001. IMTC 2001.
Proceedings of the 18th IEEE, vol. 1. IEEE, 2001, pp.
460–464.

7. D. M. Merhar, “Piezoelectric vehicle impact sensor,”

Oct. 31 1972, uS Patent 3,701,903.

8. T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D.

Ramanan, P. Doll´ar, and C. L. Zitnick, “Microsoft
coco: Common objects in context,” in European
conference on computer vision. Springer, 2014, pp.
740–755.

© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 970