Sundarapandian et al. (Eds) : ACITY, AIAA, CNSA, DPPR, NeCoM, WeST, DMS, P2PTM, VLSI – 2013
pp. 49–55, 2013. © CS & IT-CSCP 2013 DOI : 10.5121/csit.2013.3406
AUTO-MOBILE VEHICLE DIRECTION IN
ROAD TRAFFIC USING ARTIFICIAL
NEURAL NETWORKS
1B.SankaraSubramanian, 2R.Vasanth Kumar Mehta and 3N. Kumaran
1Department of CSE, SCSVMV University, kanchipuram, Tamil Nadu, India
coolsankara@gmail.com, 919486118617
2Department of CSE, SCSVMV University, kanchipuram, Tamil Nadu, India
vasanthmehta@gmail.com, 919095984004
3Department of IT, SCSVMV University, kanchipuram, Tamil Nadu, India
natarajankumn@rediffmaill.com, 919894744597
Abstract
So far Most of the current work on this area deals with traffic volume prediction during peak
hours and the reasons behind accidents only. This work presents the analysis of automobile
vehicle directing in various traffic flow conditions using Artificial neural network architecture.
Now a days, due to unprecedented increase in automobile vehicular traffic especially in metro-
Politian cities, it has become highly imperative that we must choose an optimum road route in
accordance with our requirements. The requirements are : volume of the traffic, Distance
between source and destination, no of signals in between the source and destination, the nature
of the road condition , fuel consumption and Travel Timing. Artificial Neural networks, a soft
computing technique, modeled after brain biological neuron functioning, helps to obtain the
required road way or route as per the training given to it. Here we make use of Back
propagation network, which changes the weights value of the hidden layers, thereby activation
function value which fires the neuron to get the required output.
KEYWORDS
Vehicle route selection and direction, Artificial neural network, Road Traffic Analysis, Back
Propagation network.
1. INTRODUCTION
The daily mess that a commuter faces during his travel to office is felt a lot. People get stranded
and become weird while passing through a heavy traffic condition. Though the Government plans
and ideas like diverting the traffic during peak hours, change in timing between school-college
goers and company workers, laying by-pass road and etc., to reduce the complexity in road
traffic, it has not yielded a much expected result. As this is the case, the total burden lies on the
shoulder of the commuters. They need some intelligent method by which they can avoid this
50 Computer Science & Information Technology (CS & IT)
difficult condition and make their journey a pleasant one The main constraints felt by them are :
traffic volume, no of signals they have to wait for clearance, road condition and fuel
consumption, distance they have to travel if they divert their journey in various alternative routes
to reach the destination. So, naturally, our aim is to select the best route that satisfies the user
requirements during normal working days. But in some special days, like holidays or strike call,
we can intuitively and easily judge the viable route to take up.
The paper is organized as follows. Section 1 is the Introduction. Section 2 deals with the basics of
Artificial neural networks and back propagation network. Section 3 describes about the related
work in this area. Section 4 describes the basic idea lying behind this work. and the architecture.
Section 5 discusses about the performance evaluation and result details about performance and
results. Section 6 finally concludes.
2. ARTIFICIAL NEURAL NETWORK
Artificial Neural Network [1] is a mimic of biological brain neuron for processing information
given to it.The large number of neurons interconnected among themselves work altogether to
solve a specific problem. The inputs given to it are classified into two types: Training data, which
makes the network to learn to give the required output for a given set of inputs, Testing data,
which tests the prediction accuracy of the network for randomly incoming data. It can be used for
applications such as character and digit recognization or classification of data. It has got input
layers , hidden layers and output layers. The function of the hidden layers is to fire an activation
function like sigmoidal function to bring out the required output. The weight value, assigned
between input and hidden layers changes themselves ,helps to minimize the error while getting
the desired output.
2.1 Back Propagation Network
A single layer network can accomplish very limited number of tasks. Minsky and Papert(1969)
[2] showed that a two-layer feed forward network can have many advantages but they question
remained unanswered was “ how the weights from the input to hidden layers can be adjusted?”
The answer was given by Rumelhart, Hinton and Williams in 1986. They reasoned that by
propagating errors of the output layers back into the units of the hidden layers, we can adjust the
weight values thereby making the network more adaptive and accurate . This method is also
called “Back Propagation Learning Rule”. It is a generalization of delta rules for non-linear
activation functions and multi-layer networks and a systematic method of training the networks.
A Back propagation network[2] consists of at least three layers of units:
1. An input layer
2. At least one intermediate layer called hidden layer.
3. An output layer.
The units are connected in feed-forward fashion with input units are fully connected to the hidden
layers and hidden layers are fully connected to the output layers. With back propagation
networks, learning occurs during learning phase. The patterns are applied to the input units and
propagated forward. The output layer pattern of activation is compared with the expected Pattern
to calculate the error signal. The error signal at the output pattern is then propagated back to the
input units in order to appropriately adjust the weights in each layer of the network. Once the
Computer Science & Information Technology (CS & IT) 51
back propagation network has learnt to classify correctly for a set of inputs, a second set of
inputs can be given to see how well it classify the untrained inputs. The important point is to see
how well the network generalizes.
3. THE RELATED WORK
a) Konjicija.S, Avdagic . Z and Meier.G, Wurmthaler.C in their paper [4] have shown a simple
ability of using neural networks in longitudinal vehicle guidance. Here neural networks learn
from acquired real driver data, and also the driver behaviour styles for both ends ie., extremely
comfort to extremely sportive ones. This scenario is shown as a simulated model based on
longitudinal trajectory generation. In this an adjustable comfort parameter is used for different
sorts of driver behaviour.
b) In their paper Lioing Fua and L.R.Rilettba [5] presents an artificial neural network (ANN)
based method for estimating route travel times between individual locations in an urban traffic
network.. The methodology developed in this paper assumes that route travel times are timedependent
and stochastic and their means and standard deviations need to be estimated. Three
feed-forward neural networks are developed to model the travel time behaviour during different
time periods of the day the AM peak, the PM peak, and the off-peak. These models are
subsequently trained and tested using data simulated on the road network for the City of
Edmonton, Alberta. The ANN model is compared with a distance-based model and a shortest
path algorithm. The practical implication of the ANN method is subsequently demonstrated
within a dial-a- ride para-transit vehicle routing and scheduling problem. The computational
results show that the ANN-based route travel time estimation model is appropriate, with respect
to accuracy and speed, for use in real applications.
c) Baluja, S presents [6] creation of an artificial neural network based autonomous land
vehicle controller. The performance of evolved controllers is better in unseen situations than
those trained with an error back propagation learning algorithm . An overview of the previous
connectionist based approaches is described and the exploration of methods for reducing the high
computational costs of training ANN with evolutionary algorithms is done. The evolutionary
algorithms guided by error metrics shows improved performance over those guided by the
standard sum-squared error metric. Both g evolutionary search and error back propagation are
integrated in this work.
d) In his paper, Dean A,Pomelleau [7] in The ALVINN (Autonomous Land Vehicle In a
Neural Network) project describes about real-time training of ANN to perform difficult
perception tasks. It is a back propagation network modeled to drive the CMU Navlab, a modified
Chevy van. He describes the training techniques for ALVINN to learn in under 5 minutes to
autonomously control the Navlab by observing the human driver reactions r. This enables the
ALVINN to drive in a variety of conditions including single-lane paved and unpaved roads, and
multilane lined and unlined roads, at speeds of up to 20 miles per hour.
4. THE PROPOSED WORK
We take source to destination distance (categorized as Lengthy, Medium and Short),Volume of
Traffic(categorized as Peak or Heavy, Medium and Mild), No of Signals( categorized as High,
Medium and Low), Road Condition(Categorized as Smooth, Good and Bad), Travel
52 Computer Science & Information Technology (CS & IT)
Timing(Categorized as High, Normal and Low) as input parameters as shown in fig 2(Neural
Network Back Propagation Architecture) We eliminate fuel consumption just because it is
indicated by the source and destination distance and Travel Timing selection categories. So there
are five input parameters and each have three categories, thus giving a total of 243(35)
combinations. Here we consider 5 alternative routes as the output nodes. They are :
S-X1-Y1-Z1-D,
S-W2-X2-Y2-Z2,
S-X3-Y3-Z3-D,
S-V4-W4-X4-Y4-Z4-D,
S-X5-Y5-Z5-D
where V, W, X, Y and Z intermediate nodes along the routes.
3
2
DESTINA
1
SOURCE TION
4
5
Fig.1 Main Route with Alternatives
1 Main route
2,3,4,5 Alternative routes
Signal point
So there are 15 input nodes and 5 output nodes . The hidden layers nodes are fixed as 10 , the
mean value of input and output nodes as per thumb rule. The Sigmoid function is selected as the
activation function in the hidden layer. Each input node is assigned a weight and fully connected
to the hidden layer and in turn hidden layer nodes are again fully connected to output nodes. Out
of total 243 combinations, we take any 180 as the training data sets. These inputs are uses to
train the network to produce the actual result.
For example, if we choose
Source to Destination distance as Medium (0,1,0),
Volume of Traffic as Mild (0,0,1)
No of Signals as Medium(0,1,0)
Road Condition as Smooth(1,0,0)
Travel Timing as Low(0,0,1)
Then network is trained, say, to choose Route 5 (the route with higher output layer value)
Similarly for other input conditions the network is trained and we obtain Route information as
Computer Science & Information Technology (CS & IT) 53
follows (only few samples are shown)
Now the testing data is fed into the network from unforeseen inputs. The error value , if any , in
output layer is fed back by the back propagation network and weight values of hidden layers are
changed accordingly to get the desired output. This is the usual practice for any working
day/Normal day. But for holidays or strike call we can choose always the shortest route(Default
route) to reach the destination.
Figure.2 Neural Network Back Propagation Architecture
VT – Volume of Traffic
DIS – Distance between source and destination
SIG — No of Signals
RC — Road Condition
TT – Travel Timing
H1, H2, …..H10 – Hidden layer Units
R3 — Main (Default) Route
54 Computer Science & Information Technology (CS & IT)
R1, R2, R4, R5 – Alternative routes
5. PERFORMANCE EVALUATION AND RESULT
We considered the Koyambedu-Broadway route in Chennai,TamilNadu, India as the source and
destination pair and we put one route as the normal and selected other 4 routes(left-side 2 and
right-side 2 to the normal route,see fig 1) through various road links. Initially we tested manually
about the outcome of the best route among the five alternative route chosen for various input
parameters and these outcomes are used to train the network for various combinations of input
conditions. Then the testing data are given and the results obtained are summarized as follows
Table 1, Percentage Accuracy Of Selection And Rmse
Parameters Network
tested
manually
Network tested
without Back-
Propagation
Network tested
with Back-
Propagation
%
Accuracy 95.40% 84.90% 91.20%
of Selection
Root-Mean
Square 0.28 0.42 0.12
Error
Equation : RMS for a Neural Network [3]
(1)
So we can conclude that the network with back propagation performs better than without it. The
reason behind the higher error value in case of without back propagation network is the curse of
dimensionality. Here there are 15 inputs and the categorical division(each 3) of it results in 243
combinations. The no of inputs can be reduced if we consider ANY CONDITION , say, Road
condition ANY or No of Signals ANY Numbers. Then the input reduces to 27 (33) from
243(35)and the network performs more efficiently than with higher inputs.
6. CONCLUSION
This paper presents the novel approach of neural network based analysis of best route selection
for Automobile vehicle travel and is the first of its kind. There are certain assumptions like : The
road condition remain same independent of time, the no of alternate routes remain fixed ,no of
signals are fixed, Volume of traffic remain same throughout the route during the time period
considered and the regulation over the movement of different vehicle types such as Heavy,
Medium and Light. But these assumptions are changeable and dynamic in nature. Hence the
results obtained may not hold good and give only approximate measures. In Future work, we are
proposing to analyze these dynamic nature of traffic flow conditions.
Computer Science & Information Technology (CS & IT) 55
ACKNOWLEDGEMENT
We sincerely thank, University management, for giving constant support and Mr.L.Ganesh for
his valuable help in publishing this manuscript.
References
[1] www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/ report.html :
[2] http://www.myreaders.info/03_Back_Propagation_ Network.pdf
[3] www.heatonresearch.com › … › Chapter 4: How a Mac hine Learns
[4] Konjicija, S.; Avdagic,Z.; Meier,G.; Wurmthaler, “Longitudinal vehicle guidance using neural
networks Computational Intelligence in Robotics and Automation”,. CIRA 2005.Proceedings.2005
IEEE International symposium.Page(s): 685 – 688
[5] “Estimation of Time-Dependent, Stochastic Route Travel Times Using Artificial Neural Networks”.
LIPING FUa’*and L.R.RILETTba 1Department of Civil Engineering,University of Waterloo,
Canada,; 2Department of Civil Engineering and Texas Transportation Institute, Texas A&M
University, Texas, , USA
[6] Baluja, S.Sch. “Evolution of an artificial neural network based autonomous land vehicle controller
Systems, Man, and Cybernetics”, Part B: Cybernetics, IEEE Transactions on Date of Publication:
June 1996 Author(s): of Comput. Sci., Carnegie Mellon Univ., Pittsburgh, PA Volume: 26 , Issue:
3Page(s): 450 – 463
[7] Dean A. Pomerleau, “Efficient training of artificial neural networks for autonomous navigation”
School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213 USA , Neural
Computation journal Volume 3 Issue 1, Spring 1991 Pages 88-97,MIT Press Cambridge, MA, US
Authors
B.Sankara Subramanian. B.E(EEE), M.E(CSE), (Ph.D) He was born and brought-up in
Madurai, a small town in southern Peninsular India famous for temples. He graduated
B.E(EEE) in First Class from Madurai Kamaraj University in 1997 and his Post
Graduate M.E(CSE) in 2007 from Anna University, Chennai. He has around 10 years
of teaching experience in Electrical and Computer Science Departments. His area of
interest includes Information and Database Systems, Data warehousing and Data
mining, Algorithms, Fuzzy Logic, Rough set Theory and Knowledge Retrieval. He is currently doing his
Ph.d in Soft Computing area in SCSVMV University, Kanchipuram where he is presently employed.
R.Vasanth Kumar Mehta. M.S., (Ph.d), He completed his graduate studies from BITS,
Pilani and is presently pursuing Ph.D. from Sri Chandrasekharendra Saraswathi Viswa
Mahavidyalaya(SCSVMV University), Kanchipuram, where he is working as Assistant
Professor in the Department of Computer Science and Engineering. Areas of interest
include Data Mining and Highperformance Computing
N.Kumaran did B.E(CSE),M.Tech(IT), (Ph.D). He did his B.E(Computer science and
Engineering) in CIT, Coimbatore, Tamil Nadu, India in the year 1998. And M.Tech(IT)
Sathyabama University, Chennai in 2007. Currently he is pursing Ph.D(Computer
networks) in NIT, Trichy. His areas of interest include Computer Networks, Neural
networks, Network Security and cryptography. He has got around12 years of teaching
experience various engineering colleges.
...(download the rest of the essay above)