An Intelligent Transportation System for Accident Risk Index

Quantification

Andreas Gregoriades

, Kyriacos Mouskos

and Harris Michail

Department of Computer Science and Engineering, European University Cyprus, Nicosia, Cyprus

Cyprus Transport and Logistics Ltd, Nicosia, Cyprus

Department of Electrical Engineering and Information Technology, Cyprus University of Technology, Limassol, Cyprus

Keywords: Bayesian Networks, Dynamic Traffic Assignment, Road Safety.

Abstract: Traffic phenomena are characterized by complexity and uncertainty, hence require sophisticated

information management to identify patterns relevant to safety and reliability. Traffic information systems

have emerged with the aim to ease traffic congestion and improve road safety. However, assessment of

traffic safety and congestion requires significant amount of data which in most cases is not available. This

work illustrates an approach that aims to alleviate this problem through the integration of two mature

technologies namely, simulation-based Dynamic Traffic Assignment (DTA) and Bayesian Networks (BN).

The former generates traffic flow data, utilised by a BN model that quantifies accident risk. Traffic flow

data is used to assess the accident risk index per road section and hence, escape from the limitation of

traditional approaches that use only accident frequencies to quantify accident risk. The development of the

BN model combines historical accident records obtained from the Cyprus police and domain knowledge

from road safety.

1 INTRODUCTION

Road safety constitutes a problem of paramount

importance worldwide (Bartley, 2008). To deal with

this problem, intelligent transportation systems (ITS)

have emerged. ITS are also used in the following

areas: congestion control, mobility enhancement,

delivering environmental benefits, and boosting

productivity and expanding economic and

employment growth. The work presented herein

describes a novel approach and tool for assessing the

accident risk index of road networks. This

prerequisites the assessment of accident risk.

According to (Zheng, 2009), accident risk models

are divided into two categories: social risk models,

that measure probabilistic (frequentist) collective

damage, and individual risk models, that measure

probabilistic (frequentist) individual damage. These

are categorized into aggregate and disaggregate

methods. The former, use global statistics and the

former specific events (Bartley 2008). However,

predicting accident risk requires not only

frequencies of crashes per road section but also

traffic flow data. However, in most cases traffic flow

and accident data cannot be found together. To that

end authorities perform safety analysis using only

crash data which is an approximate approach to

accident risk estimation. This paper aims to address

this problem through the development of a novel

Intelligent Traffic Information System (ITIS) that

leverages the capabilities of two mature

methodologies namely simulation-based Dynamic

Traffic Assignment (DTA) embedded in the VISTA

simulator (Ziliaskopoulos et al., 1996) and Bayesian

Belief Networks (BN). The former is widely used in

transportation planning and operations to predict

drivers’ decisions (where and when to travel on the

road network), and in work was used to estimate

traffic flow conditions for each road section. The

latter is a powerful uncertainty modelling technique

used for the quantification of accident risk under

varying conditions.

The paper is organised as follows. Next section

describes the methodology. Subsequent sections

concentrate on data pre-processing and BN model

development. The integration of VISTA with the BN

along with the results that emerge from the

amalgamation of the two technologies in an ITS, is

described next. The paper finishes with conclusions.

318

Gregoriades A., Mouskos K. and Michail H..

An Intelligent Transportation System for Accident Risk Index Quantiﬁcation.

DOI: 10.5220/0003989203180321

In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS-2012), pages 318-321

ISBN: 978-989-8565-10-5

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 METHODOLOGY

The Road Safety Assessor, ITS system proposed

herein is the amalgamation of probabilistic risk

assessment with a mesoscopic traffic simulation,

namely VISTA. The need for this integration boils

down to the limitations of traditional traffic

information systems that mainly concentrate of data

warehousing. The methodology proposed utilises

data marts to generate projections of future system

behaviour. To that end, intelligent information

management techniques are employed to distil

knowledge used to develop models that enable the

prospective system behaviour. The two models that

emerged from this process are the accident risk

assessment model and the traffic simulation model.

The accident risk assessment employed is causality-

based and uses BN. In BN each node is used to

represent a random variable that has been identified

to have a causal influence on accident risk. Each

directed edge represents an immediate dependence

or direct influence between parent and child

variables (Jensen, 2001). Evidence is entered in the

model through instantiation of leaf node on the

model. Inference is achieved by belief propagation

through the models topology. BN technology is used

to model how traffic and infrastructural factors

influence accident risk. The second component of

the approach is a road traffic simulator based on

DTA. The DTA model is used in VISTA through the

Dynamic User Equilibrium (DUE) model (Peeta et

al., 2000). The use of DTA model enhances the

limitations of existing practices by providing a

consistent way of producing estimates of traffic flow

conditions of road networks using limited

information from traffic flow detectors. Moreover, it

produces timely and complete traffic volume

estimates for all sections of a road network and

hence, can be used to assess accident risk using time

varying conditions. The integration of BN with

VISTA in the proposed traffic information system

enables the dynamic assessment of accident risk

using simulated traffic conditions and prior

knowledge embedded in the BN. A pilot study

conducted with the system aimed to assess the safety

performance of the Nicosia road network in Cyprus

and to investigate how it will behave under different

scenarios.

Initially the road traffic model of Nicosia was

specified, implemented, verified and validated in

VISTA. Models in VISTA are represented by nodes

connected by unidirectional links that represent flow

of traffic in one direction. It is possible to have more

than one link between two nodes to indicate separate

lanes and lane direction. The completed VISTA

simulation model was integrated with an accident

risk assessor implemented in Java. The simulator

provided the risk assessor with the traffic volumes of

all road sections of the network for every 15 min

interval. Traffic volumes along with infrastructural

properties of the network were used by the BN to

assess accident risk on a simulation step basis. For

the development of the BN topology and the

parameterization of its prior knowledge, historical

road accident data were utilized.

3 ARCHITECTURE OF THE ITS

The Road Safety Assessor tool emerged from the

integration of VISTA with BN technologies. The

main components of the tool are: the BN engine, the

accident risk assessor, the VISTA simulator, the data

pre-processor that incorporates the scenario

generator, the results analyzer and the visualizer.

The tool was developed using a component-based

software engineering methodology. With the initial

specification of the system requirements captured,

we proceeded in the identification of suitable

software components that matched the initial system

requirements. These components were subsequently

integrated to implement parts of the system’s

functionality. In particular the Bayesian inference

engine and the visualization components were

selected after thorough investigation. The glue-code

that enabled components integration was

implemented in Java. The risk assessor quantifies

accident risk using a Bayesian inference engine that

utilizes the probabilistic model of accident risks.

Input to the BN assessor is categorized into static

and dynamic. The former is obtained from the

VISTA database and the latter is the output of the

VISTA simulation.

Input to the accident risk assessor is organized in

the form of scenarios. An input scenario to the BN

assessor is defined by the static and dynamic

properties of each road section. Static information is

obtained from the VISTA database and in

combination with the dynamic input from the

simulator. This provides the baseline for generating

a number of plausible test scenario variations for

each road section. Generated scenarios are executed

by the risk assessor to quantify the probability of

accident. The scenario generator is responsible for

generating plausible scenario variations to stress-test

the safety performance of each road section. The

visualizer processes the results and depicts these to

the user graphically. Input scenarios are executed by

AnIntelligentTransportationSystemforAccidentRiskIndexQuantification

319

the BN model. Each scenario evidence is propagated

down the BN topology to produce the posterior

probability of accident risk per scenario.

The integration of the VISTA with the BN model

was realized through asynchronous data interchange.

To establish communication between VISTA and

the risk assessor it was imperative to pre-process

VISTA’s output data prior to being utilized by the

BN in the risk assessor. Specifically, VISTA

variables are continuous by nature, hence, had to be

converted into categorical/discrete to be processed

by the BN model, since it uses only discrete nodes.

Hence, it was necessary to discretize the output from

VISTA prior to instantiating the BN model. For the

discretization process it was necessary to refer to

domain experts that specified the cut-off values for

each variable. Specifically, for traffic volume three

states were defined, namely, low, average and high.

The first corresponding to less than 100 vehicles per

15 time interval, the second to less than 350 and the

last to greater than 350.

4 BN MODEL DEVELOPMENT

Development of BNs requires the specification of

the topology and the conditional probability tables.

To that end historical accident records were obtained

from the traffic safety department of the Cyprus

Police. Preliminary compilation of the data was

performed with the SPSS statistical package to

reduce the dimensionality of the data. The accident

dataset covered all accidents occurred in the Nicosia

area from 2002 until 2008 and comprised over 9000

records. Each record consisted of 43 (six continuous

and 37 categorical) input parameters covering

global, local, temporal, accident, driver and car

characteristics collected at the site of the accident by

the police officers, eye witnesses and the involved

parties. Each record was associated with a single

categorical output parameter pertaining to accident

severity, namely light, severe and fatal, as evaluated

by the police officer at the site of the accident.

However, for the development of the BN model

topology it was imperative to enhance the dataset

with additional information regarding the traffic

conditions of each accident record from VISTA

simulation. Therefore each accident record was

mapped on a geospatial GIS platform and

subsequently import on VISTA to obtain the

dynamic information of each accident location at

different time intervals. This yielded an enhanced

dataset of accident records.

Figure 1: Data inferred BN topology at execution time.

A preliminary analysis of the dataset provided a

generic indication of the influence of each variable

to road accident risk. Data pre-processing was

performed in two steps (a) replacement of missing

and erroneous parameter values by the mean value,

and (b) grouping related values of multi-valued

categorical parameters so as to have a manageable

number of states per parameter. Next, to reduce the

dimensionality of the dataset, Principal Component

Analysis (PCA) was used. This helped to identify

the core variables of the model. Results from the

dimensionality reduction using PCA, yielded 19

variables for the BN topology. The topology

depicted in Figure 1, was learned from processed

dataset using the Expectation Maximisation

algorithm (Jensen, 2001). Figure 1 also shows an

instantiation of the BN model in Hugin researcher

tool. The developed ITS utilises the Hugin engine

using its API. Each variable in this figure is

accompanied by a monitor window that shows its

states. The input evidence is showed as a solid bar in

the monitor window of each variable. Collectively

all variable instantiations correspond to one scenario

variation that is provided by the scenario generator

component of the tool that uses input from VISTA.

In each scenario variation variables that are not

instantiated using input from VISTA is varied

systematically to produce additional scenario

permutations that instantiates the BN model.

To estimate the accuracy of the developed BN

model, validation was performed using the accident

dataset obtained from the police. The dataset was

utilised to identify locations on the network with

high accident frequency. These are the networks

ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems

320

black spots. These points were used to validate the

model after it was implemented. Specifically, a

subset of the accident dataset was used to validate

the system. Black-spots that were identified using

the dataset, were used to test the BN accuracy under

varying conditions of traffic flow data.

5 RESULTS

Results from the accident risk assessor were used to

calculate the accident risk index (ARI) of each road

section. BN scenarios for each road segment were

labeled accident prone if the BN accident risk

probability was above a pre-specified threshold

value. BN scenarios that fell below the threshold

value were ignored. Scenarios were defined on the

fly by the scenario generator component. Each

segment is evaluated against scenarios that describe

traffic condition at different time intervals and driver

profiles. To assess the ARI it was imperative to

normalize the number of accidents that were

predicted by the BN with the traffic volume per time

interval, for each road section. To that end, the

developed system uses a systematic approach that

utilizes the traffic volume estimates from the VISTA

simulation and the accidents predicted using the BN

risk assessor. Traffic volume acts as a normalizing

factor for the number of accidents predicted using

the BN risk assessor. In this study, the ARI is

defined as:

Accident Risk Index (ARI) = Number of

accidents predicted by the BN/estimated

traffic flow rate per time period of the day,

from DTA

ARI results gave rise to road sections that inherently

have safety issues. These are the network's black

spots. An illustration of the preliminary results

produced by the method is depicted in Figure 2. This

figure illustrates a subset of the results and indicates

that sections with IDs, 3, 21 and 47 have the highest

ARI.

6 CONCLUSIONS

The ITS system described herein illustrates a novel

approach to quantifying road safety using

probabilistic inference expressed in causal

relationships between factors leading to accidents

with DTA simulation. The method escapes from the

problem of traffic data shortage through the use of

DTA simulation. VISTA provides complete traffic

volume data estimates for all road sections of the

network on a 24 hour basis. This constitutes

advancement over existing methods that base their

analysis on limited data obtained from a scarce

number of traffic sensors on the network.

Figure 2: All road section with their ARI values (Y axis).

REFERENCES

Bartley, P., 2008. Traffic Accidents: Causes and

Outcomes. Nova.

Florian, M., Mahut, M., Tremblay, N., 2008. Application

of a simulation-based dynamic traffic assignment

model. European Journal of Operational Research,

189, 1381–1392.

Jensen, F., 2001. Bayesian Networks and Decision

Graphs. Springer.

Peeta, S., Ziliaskopoulos, A., 2001. Foundations of

Dynamic Traffic Assignment: The Past, the Present

and the Future. Networks and Spatial Economics, 1

(3/4), 233-65.

Zheng, X., Liu, M., 2009. An overview of accident

forecasting methodologies. Journal of Loss Prevention

in the Process Industries, 22(4), 484-491.

Ziliaskopoulos, A., Lee, S., 1996. A Cell Transmission

Based Assignment-simulation Model for Integrated

Freeway/Surface Street Systems. Proc., 75th

Transportation Research Board, Annual Meeting,

Washington, DC.

AnIntelligentTransportationSystemforAccidentRiskIndexQuantification

321