Information Technology Reference
In-Depth Information
infrastructure and includes information such as the speed of vehicles and the number
of vehicles in a given amount of time on a given length of road. Traffic flow parame-
ters are measured using induction loop sensors that are embedded at regular distances
along many major roads.
In this research work, we describe a case-based reasoning (CBR) system that ana-
lyses current traffic flow data obtained from road sensors and makes a prediction
regarding accidents by comparing current flow data with historical flow data that
preceded an accident in the past. CBR is an artificial intelligence methodology that
solves problems based on similar problems in the past [3]. The advantage of CBR is
that it does not depend on patterns that are analysed or defined a priori and can be
used to infer solutions, when problem domains are not well understood or are difficult
to describe using mathematical formulations or rules. Since CBR does not depend on
well-defined rules, interactive and derived factors are inherently taken into account.
Traffic flow patterns, such as congestion or traffic waves, which make accidents more
likely, are indirectly considered in the flow data contained in the archived accident
cases. Further, as the sensors record flow data at regular intervals of time and dis-
tance, the data is spatio-temporal in nature. So far, most traffic flow forecasting me-
thods described in the literature considered only sensor data obtained from a single
sensor over an interval of time (temporal data) [4-7].
The novelty of the research work presented lies in the application of CBR to identi-
fy potentially hazardous traffic flow conditions based on past similar traffic flow con-
ditions considering not only temporal but both spatial and temporal flow data.
The first step in determining the feasibility of using CBR to predict when traffic
flow conditions might cause an accident is to assess if a CBR system is capable of
differentiating between traffic flow data that was followed by an accident and generic
traffic flow data that was not followed by an accident. The experiments presented in
this paper focus on this aspect.
The paper is organised as follows. A brief overview of the background to the prob-
lem domain and relevant literature is given in Section 2. The setup of the system to
investigate the ability of CBR to differentiate between accident and non-accident
traffic flow data is presented in Section 3. Section 4 describes experiments and
presents preliminary results. The salient points of our research so far are summarized
in Section 5, along with future research directions.
2
Background and Related Work
This section briefly introduces the fundamental background knowledge required for
the comprehension of the research work. Previous and related relevant work from the
literature is included. The induction loop sensors that measure traffic flow are de-
scribed followed by a brief introduction to case-based reasoning.
2.1
Traffic Flow Data and Measurement
Traffic flow data is measured by induction loop sensors embedded in roads at regular
intervals, which form part of a distributed network called the Motorway Incident
Search WWH ::




Custom Search