Smart real-time traffic congestion estimation and

clustering technique for urban vehicular roads

Vishwajeet Pattanaik,
Mayank Singh
Computer Science and Engineering Deptt.,
Krishna Engineering College
Ghaziabad, India 201 007
Email: vpattanaik@gmail.com,
mayanksingh2005@gmail.com
P.K. Gupta
Department of Electrical, Electronic,
and Computer Engineering
University of Pretoria,
South Africa, 0028
Email: pkgupta@ieee.org
S.K. Singh
Department of Computer Science
& Engineering
Indian Institute of Technology BHU,
Varanasi, India, 221005
Email: sks.cse@iitbhu.ac.in

Abstract—Road-traffic congestion is becoming a serious con-
cern in developing countries and impacts the economy of coun-
tries gravely. Increasing congestion on urban roads presents a
genuine threat to the economic growth and livability of city
regions. Most traffic congestions are caused due to unplanned
road networks, high volumes of vehicles and presence of critical
congestion areas. Traffic congestions not only pose a threat to
the economy but also to the environment. Spillover effect from
congested main roads to secondary roads and side streets as
alternative routes often leads to more congestion; increasing
the chances of collisions and accidents due to tight spacing
and constant stopping-and-going. The following paper presents a
smart congestion avoidance technique by estimating the scope
of real-time traffic congestion on urban road networks and
predicts an alternate shortest route to the destination. The
proposed system uses K-Means Clustering Algorithm to estimate
the magnitude of congestion on different roads and then employs
Dijkstra’s Algorithm to predict the shortest route. Once the user
inputs the destination into the system, the system predicts the
shortest route from the user’s current location. The process is
reiterated at every intersection until user reaches the destination.
Index Terms—Congestion avoidance, K-Means Clustering,
Path Planning, Shortest Path, Neighborhood graph
I. I NTRODUCTION
Urbanization in developing countries like India and China
has led to massive population burst in their metro cities
like New Delhi and Beijing. Mobility in these cities or the
transport network within cities tremendously impacts both the
city’s and the nation’s socio economic growth [1]. In India,
urban population is growing at an average rate of around
3% per year. According to survey conducted this growth will
touch the figure 500 million from existing 377 million by the
year 2021 in urban areas [2]. With such a rapid growth in
urban population, there has been an increase in demand for
mobility and with it an increase in vehicle population on roads.
Due to increase in vehicles on roads, traffic congestion has
become a severe problem in urban areas. Traffic congestion
constrains the growth of Gross Domestic Product (GDP) of any
developing country. Due to traffic congestion various problem
arise such as increase in air pollution, vehicle operating costs,
traveling time, etc. According to a report [3], Rs. 600 billion
are lost every year due to traffic jams or delays on high
volume roads and highways which also includes fuel wastage.
Numerous actions and policies have been put into place over
the years to reduce these traffic congestions, but no reasonable
impacts of the same have been seen yet. Hence, it is obligatory
that systems and algorithms be designed in such a way that
people are able to avoid the traffic congestions in real time.
One can use applications or services to navigate within cities,
most of which suggest two to three route options along
with various details like average speed, current congestion
segments, tolls, distances,constructions, etc. Most of these
applications or services work on previously identified routes
which may or may not suffice to current scenarios. As per real
time congestions are concerned, most applications or services
suggest only present congestion situations on suggested routes
[4] [5]. It often so happens that by the time a user reaches the
previously predicted congested area, the congestion is already
cleared out. In almost all cases once users select a route,
the software only guides him through the chosen route. None
of the applications or services reroute the user according to
real time traffic congestions as the trip progresses. Existing
applications or services uses various techniques to estimate the
traffic on road such as fuzzy control theory, microscopic traffic
simulations, etc.; all of which are extremely expensive and
require specialized hardware. In this paper, proposed traffic es-
timation system is based on Global Positioning System (GPS)
enabled mobile phones which are abundantly available making
the system particularly suitable for developing countries. In
view of the above mentioned problems, a real time congestion
avoidance method is proposed which highlights a method
for alternate shortest route prediction based on K-Means
Clustering of traffic data points. The proposed methodology
can predict which road segments are congested or cleared
through real time GPS data. The system informs the driver
about real time traffic conditions and adjusts the route so as
to avoid congestions and reduce traveling time drastically.
II. R ELATED W ORK
In [6], Kanoulas et al. have proposed a method for finding
the fastest path through a road network.This method was

978-1-5090-2597-8/16/$31.00 2016
c IEEE 3420
based on the constraint of trip time interval at either the
start of or destination. In [7], Moosavi and Hovestadt have
proposed a framework for conceptual driven traffic data.This
framework was based on application of Markov chain with
data stream from GPS data on taxis. To estimate the traveling
time, sample data from GPS equipped vehicles are collected.
In [8], Antoniou et al. have described a methodology for
prediction of short term traffic state andwas comprised of
components such as nearest neighbour classification, model
based clustering, and variable length Markov chains. In [9],
Yan Qi has presented probabilistic models for short term traffic
conditions predictions. This model was also compared with
the traffic prediction using HMM based model. In this paper,
author has derived the traffic features from embedded magnetic
loop in the road. In [10], Asahara et al. have proposed a state
space modelling method for predicting pedestrian movement
on the basis of mixed Markov chain model. Karbassi and Barth
[11] have proposed a car sharing application with defined
start and end point. Liu and Bai [12] have achieved vehicle
location tracking on Internet by using GPRS vehicle location
terminals. They have also combined the identified location
with Internet technology and GIS technology. In [13], Al-
Taee et al. have presented a distributed system for remote
monitoring of vehicle diagnostics and geographical position. In
[14], Rajput et al. proposed an algorithmic framework which
combines the rough set theory with k-means to obtain the
appropriate clusters.
Each of these proposed papers suggested various techniques
for Path Finding on Road Networks; most of them using
latest algorithms and practices including GPS, GPRS and
IoT (Internet of Things). Unfortunately none of the proposed
methods consider traffic congestion as a constraint while
finding of navigation path. Proposed technique in this paper,
fixes this issue using smart real-time clustering of traffic by
k-means clustering algorithm. The system further weighs the
traffic clusters and then finds routes on this new weighted road
map.
III. P ROPOSED M ETHODOLOGY
The proposed system’s intension is to identify the shortest
path from the driver’s current location (source) to his/her
destination avoiding traffic congestions in the best possible
fashion. To accomplish this objective, the proposed system
identifies road networks of a particular area from Google
Maps. The system then converts the road network into a
weighted graph where each intersection is denoted by a node
and each edge represents a road. Based on weighted graph, a
neighborhood metric is generated with all intersecting nodes;
further applying Dijkstra’s algorithm to discover the shortest
path (with minimum traffic congestion) from source to des-
tination. The proposed real time traffic congestion avoidance
system as shown in Algorithm 1, is primarily composed of
five basic steps. 1) Fetch road map of the driver’s vicinity.
Assume the driver’s current location is the source, and ask
the user to feed in his/her destination. 2) Identify the shortest
path from source to destination in ideal condition i.e., without
2016 IEEE Region 10 Conference (TENCON) — Proceedings of the International Conference
traffic. 3) Collect real time GPS data of vehicles (through
GPS based mobile phone app) and plot the real time traffic
data onto the road map. 4) Use the real time traffic data
attained in the previous step to create traffic clusters using K-
Means Clustering. 5) Formulate the density of traffic clusters
and identify the alternate shortest routes to avoid congestion.
Reiterate the process in real time until the driver reaches to
destination.
Algorithm 1 Congetion Avoidance Algorithm
1: Fetch Driver’s Current Location (Source)
2: Get Destination Coordinates
3: Retrive Road Map of Area
4: while Source = Destination do
5: Retrive Real Time Traffic Data from App
6: Plot Traffic Data onto 2D Problem Space
7: Apply K-Means Clustering on Traffic Data
8: Assign Weights to Traffic Clusters
9: Combine Traffic Cluster Data with Road Map
10: Convert Weighted Road Map into Neighborhood Matrix
11: Apply Dijkstra’s Algorithm
12: Display Shortest Path
13: Fetch Driver’s Current Location (Source) Again
14: end while
IV. ROAD MAP TO G RAPH CONVERSION
The system fetches the road map of driver’s local area
from sources like Google Maps, Open Street Map, etc. The
labeled map is processed using image processing techniques
to fetch road map data only as shown in Fig. 1. The system
further converts the road map image (n ∗ n matrix) into
neighborhood matrix (n2 ∗ n2 matrix) where each element
represents relationship with its neighboring pixels. This matrix
represent whether there is a direct connected road is available
or not. Weights of all directly connected nodes are initialized
to 1 in the initial road network because different vehicles
may pass through these road segments. The road network
can be represented as a weighted graph G = (V, E), where
each intersection is denoted by a node V and each edge E
represent a road in graph. When a vehicle is driven from
a source S to destination D, an ordered set of roads in the
route is defined as Ai = (a1 , a2 , a3 ...an ) where Ai represents
the ith road and n is total number of roads. According
to graph theory, every route can be represented as Ai =
(e1 → v1 , v2 ; e2 → v2 , v3 ; ...; en → vn , vn+1 ).
V. T RAFFIC CLUSTERING AND CONGESTION ESTIMATION
To get the current traffic condition of driver’s local area,
the system uses real-time GPS tracking of vehicle users using
Android Application. The GPS location data transmitted by
the vehicles in collected in local clouds. The real-time data
transmitted by the App includes: vehicle location (longitude
and latitude), speed, drive time, direction, etc. An institution
based proof of concept Android App was developed to do
the same, although there are a few alternative applications or
3421

(a) (b)
(c)
Fig. 1. Conversion of Google Road Map from (a) Labelled Map to (b)
Unlabelled Map to (c)Black White 2D Matrix
services available on the internet which can be exploited to
accomplish the above mentioned tasks.
The GPS data with vehicles locations received from the
application is fed into a 2-Dimensional problem space of
size n ∗ n. The vehicle data is plotted on to driver’s local
road map road. K-means clustering is then implemented so
as to create traffic clusters with minimum Euclidean distance.
Higher number of vehicle points within a cluster would then
denote higher density, or more congestion.
Based on similarity or dissimilarity metric, clustering in N -
dimensional Euclidean space is the process of partitioning
a given set of N points into m clusters. Let, a set S is
representing n points (n1 , n2 , n3 , ...nn ) and m clusters be
representing (a1 , a2 , a3 , ...am ) Then,
Ai = φ f or (i = 1, 2, 3, ..., m)
Ai ∩ Aj f or (i = 1, 2, 3, ..., m);
(j = 1, 2, 3, ..., m) and i = j and
m

Ai , Ai = S (1)
i,j=1
The road network is represented as neighborhood matrix where
each intersection is represented by a node and each edge
represents a road segment. The initial weights are assumed to
be 1 if two vertices’s are directly connected with each other
which represents the estimated travel time. When the GPS data
3422 2016 IEEE Region 10 Conference (TENCON) — Proceedings of the International Conference
is plotted on road segments to show the availability of vehicles
on a particular road, the weights are updated periodically
according to every vehicle available on each road segment.
These updated weights are used to estimate the travel time or,
to calculate the shortest path from current location to desired
destination with congestion avoidance. To estimate the travel
time Greenshield’s model is used. The model considers that
there is a linear relationship between vehicle per meter traffic
density Di on road i and estimated road speed Si . One can
formulate this situation according to following formula,
Di
Si = Sj (1 − ) (2)
Dj
Where Dj is traffic density in jam and,Sj represents the speed
flow on i road segment
Pi
Ai = (3)
Si
Where Ai is the estimated travel time and Pi is the path length
of the same road segment. Road density is calculated with
Convex Hull algorithm on each identified k-means cluster.
Data set of vehicles available on road is provided to the
algorithm to create convex hulls. Convex hulls represent the
traffic congestion on particular road segment as shown in Fig.
2.
Now with the help of Dijkstra’s algorithm, the system
finds alternate shortest path Pi to avoid the congested roads
and to save traveling time. During the alternate path finding,
road segments with higher weight in neighborhood matrix are
ignored to avoid the congestion and estimate the new travel
time with alternate path.
Pi
Ai = (4)
Si
(a) (b)
Fig. 2. Red, Green and Blue points on (a) Black & White Road Map represent
driver’s location, destination and other vehicles, respectively (b) Visualization
of Convex hulls for traffic clusters
VI. F INDING SHORTEST PATH
The driver’s current location is then assumed to be the
source and the destination coordinates are fed in by the user.
The systems main focus is to track the driving path with actual
location and traffic conditions after periodic time, so that traffic
conditions can be monitored in real time and alternate routes
can be suggested to avoid the congestion. In this situation,
the initial source will change to actual location but destination
will be same. Alternate shortest route is provided from current
location to desired destination. In this paper, we have used
Dijkstra’s algorithm to find the shortest path from user’s source
location to its destination.
After applying this algorithm on initial neighborhood met-
ric, the shortest path from desired source to destination and
total estimated traveling time is predicted. This traveling time
is based on smooth road segments. Fig. 3 shows predicted
routes from different positions p1 , pr , and ps . As it can be seen
in the figure the routes predicted by the system are dynamic
and changes with traffic density on various roads from source
to destination.
(a) (b)
(c)
Fig. 3. Red, Green, Blue and Cyan points on Road Map represent driver’s
location, destination, other vehicles and predicted path, respectively. The path
changes dynamically with driver’s location, starting from (a) position p1 to
(b) position pr , and (c) position ps .
VII. C ONCLUSION AND F UTURE S COPE
The proposed methodology was tested over several road
maps of New Delhi. Simulation results clearly indicate drastic
reduction in travel time, compared to path finding methods
using conventional techniques. Implementing the proposed
system in Navigation systems could help to reduce economic
2016 IEEE Region 10 Conference (TENCON) — Proceedings of the International Conference
and environmental losses caused due to congestion. Also
since the system is based on real-time traffic and dynamically
predicts routes as driver’s progress towards their destination,
there is minimal chance that people starting from the source
to the same destination would ever cause congestion for
other drivers. The proposed system is designed as a proof
of concept, hence it is based on simple algorithms like:
k-means clustering, convex hull and Dijkstra’s Algorithm.
The system has high time-space complexity, which could
be reduced significantly by use of latest clustering and path
finding algorithm. Furthermore, implementing the system over
Disturbed Computing Systems could enhance the system’s
performance. The proposed system is just a step towards
radicalizing cheap yet reliable traffic congestion detection and
avoidance systems, helping save economy, environment, and
health.
R EFERENCES
[1] Dorina Pojani and Dominic Stead, ”Sustainable Urban Transport in the
Developing World: Beyond Megacities,” Sustainability, vol. 7, no.7,
pp.7784-7805, 2015.
[2] Sanjay Kumar Singh, ”Urban Transport in India: Issues, Challenges,
and the Way Forward,” European Transport/TrasportiEuropei , Issue
52, pp.27-52, 2012.
[3] Subrata Mitra,”Operational Efficiency of Freight Transportation by Road
in India,” TCI-IIMC Joint study report, pp.1-36, 2012.
[4] Yuan, Jing, Yu Zheng, Xing Xie, and Guangzhong Sun, ”Driving with
Knowledge from Physical World,” Proc. of the 17th ACM SIGKDD
International Conference on Knowledge Discovery and Data mining,
2011, pp.316-324.
[5] Leontiadis, Ilias, Gustavo Marfia, David Mack, Giovanni Pau, Cecilia
Mascolo, and Mario Gerla, ”On the Effectiveness of an Opportunitiec
Traffic Management System for Vehicular Network,” IEEE Transaction
on Intelligent Transportation System, vol. 12, no.4, pp.1537-1548, 2011.
[6] Kanoulas, E., Du, Y., Xia, T. and Zhang, D.,”Finding fastest paths on
a road network with speed patterns,” Proc. of the 22nd International
Conference on Data Engineering (ICDE’06), 2006, pp.10-10.
[7] Moosavi, V. and Hovestadt, L.,”Modeling Urban Traffic Dynamics in
Coexistene with Urban Data Stream,” Proc. of the 2nd ACM SIGKDD
International Workshop on Urban Computing, 2013, p.10.
[8] Antoniou, C., Koutsopoulos, H.N. and Yannis, G.,”Traffic state pre-
diction using Markov chain Model,” Proc. of the European Control
Conference, July, 2007, pp.2428-2435.
[9] Yan Qi, ”Probabilistic Models for Short Term Traffic Conditions,”
Louisiana State University, 2010.
[10] Asahara, A., Maruyama, K., Sato, A. and Seto, K, ”Pedestrian-movement
prediction based on mixed Markov-chain model,” Proc. of the 19th
ACM SIGSPATIAL International Conference on Advances in Geographic
Information Systems, Nov, 2011, pp. 25-33.
[11] Karbassi, A. and Barth, M., ”Vehicle Route Prediction and Time of
Arrival Estimation Techniques for Improved Transportation System
Management,” Proc. of the Intelligent Vehicles Symposium, 2003, pp.
511-516.
[12] Liu, Y. and Bai, B., ”Research on GPRS vehicle location network service
system,” Proc. of International Conference on Computer, Mechatronics,
Control and Electronic Engineering, vol.4, 2010, pp.401-404.
[13] Al-Taee, M.A., Khader, O.B. and Al-Saber, N.A., ”Remote Monitoring
of Vehicle Diagnostics and Location Using a Smart Box with Global Po-
sitioning System and General Packet Radio Service,” Proc. of IEEE/ACS
International Conference on Computer Systems and Applications, 2007,
pp.385-388.
[14] Rajput, D., Singh, P. and Bhattacharya, M., ”An Efficient and Generic
Hybrid Framework for High Dimensional Data Clustering,” Proc.of
International Conference on Data Mining and Knowledge Engineering
(ICDMKE 2010), World Academy of Science, Engineering and Tech-
nology, Rome, Apr, 2010, pp.174-179.
3423