Document downloaded from:
This paper must be cited as:
The final publication is available at
Copyright
http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6477047
http://hdl.handle.net/10251/38164
Institute of Electrical and Electronics Engineers (IEEE)
Fogue, M.; Garrido, P.; Martínez, FJ.; Cano Escribá, JC.; Tavares De Araujo Cesariny
Calafate, CM.; Manzoni, P. (2014). A system for automatic notification and severity
estimation of automotive accidents. IEEE Transactions on Mobile Computing. 13(5):948-
963. doi:10.1109/TMC.2013.35.
A System for Automatic Notification and Severity
Estimation of Automotive Accidents
Manuel Fogue, Piedad Garrido, Francisco J. Martinez
University of Zaragoza, Spain
Email: {m.fogue, piedad, f.martinez} @unizar.es
Juan-Carlos Cano, Carlos T. Calafate, Pietro Manzo ni
Universitat Politècnica de València, Spain
Email: {jucano, calafate, p manzoni}@disca.upv.es
Abstract
New communication technologie s integrated into modern vehicles offer an opportunity for better
assistance to people injured in traffic accidents. Recent studies show how c ommunication capabilities
should be supporte d by artificial intelligence systems capable of automating many of the decisions
to be taken by emergency services, thereby adapting the rescue r esources to the severity of the
accident and reducing assistance time. To improve the overall rescue process, a fast and accurate
estimation of the severity of the accident represent a key point to help the emergency services to
better estimate the required resources. This paper proposes a novel intelligent system which is able
to automatically detect road a ccidents, notify them through vehicular networks, and estimate their
severity based on the concept of data min ing and knowledge inference. Our system c onsiders the most
relevant variables that can characterize the severity of the accidents (variables such as the vehicle
speed, the type of vehicles involved, the impa c t speed, and the status of the airbag). Results show
that a complete Knowledge Discovery in Da ta bases (KDD) process, with an adequate selection of
relevant fe a tures, allows generating estimation models able to predict the severity of new accidents.
We develop a prototy pe of our system based on off-the-shelf devices, and validate it at the Applus+
IDIADA Automotive Resear c h Corp oration facilities, showing tha t our system can notably reduce
the time needed to alert and deploy the emergency services after an accident takes place.
Index Terms
KDD; Data Mining ; Vehicular Networks; Traffic Accidents Assistance
I. INTRODUCTION
During the last decades, the total number of vehicles in our roads has experienced a remarkable
growth, making traffic density higher an d increasing the drivers’ attention requirements. The imme-
diate effect of this situation is the dramatic inc rease of traffic accidents on the road, representing
1
a serious problem in most cou ntries. As an example, 2,478 people died in Spanish roads in 2010,
which means one death for every 18,5 51 inha bitants [1], and 34,500 people in the whole European
Union died as a result of a traffic accident in 2009 [2].
To reduce the number of road fatalities, vehicular networks will play an increasing role in the
Intelligent Transportation Systems (ITS) area. Mos t ITS applications, such as road safety, fleet
management, and navigation, will rely on data exchanged between the vehicle and the roadside
infrastructure (V2I), or even directly between vehicles (V2V) [3]. The integration of sensoring
capabilities on -board of vehicles, along with peer-to-peer mobile commun ic ation among vehicles,
forecast sign ificant improvements in terms of safety in the near future.
Before arriving to the zero accident objective on the long term, a fast and efficient rescue operation
during the hour following a traffic ac cident (the so-called Go lden Hour [4]) significa ntly increases
the probability of survival of the injured, and reduces the injury severity. Hence, to maximize the
benefits of using communication systems between vehicles, the infras tructure s hould be supported by
intelligent systems capable of estimating the severity of accidents, and automatically deploying the
actions required, thereby reducing the time n eeded to assist injured passengers. Many of the manual
decisions taken nowadays by emergency services are based on incomplete or inaccurate data, which
may be replaced by automatic systems that adapt to the specific characteristics of each accident.
A preliminary asse ssment of the severity of the accident will help emergency services to adapt the
human and material resources to the conditions of the accident, with the consequent as sistance quality
improvement [5].
In this p aper, we take advantage of the use of vehicular networks to collect precise information about
road accidents that is then used to estimate the severity of the collision. We propose an estimation
based on data mining classification algorithms, trained using historical data about previou s accidents.
Our proposal d oes not focus on directly reduc ing the number of accidents, but on improving post-
collision assistance.
The rest of the paper is organized a s follows: Sec tion II presents the architecture of our proposed
automatic system to improve accident assistance. Sections III, IV, and V provide details of our
Knowledge Discovery in Databases (KDD) model adapted to the traffic acc idents domain. Section VI
presents the implemented prototype built to test o ur system and evaluates the obtained res ults of
the validation process. Section VII reviews the related work on the improvement of traffic safety
through telecommunic ation technologies, and da ta mining for accident severity estimation. Finally,
Section VIII concludes this paper.
2
Fig. 1. Architecture of our proposed system for automatic accident notification and assistance using vehicular networks.
II. OUR PROPOSAL
Our approach collects information available when a traffic accident occurs, which is captured by
sensors installed on -board the vehicles. The data collected are structured in a pac ket, and forwarded
to a remote Control U nit through a combination of V2V and V2I wireless communication. Based on
this information, our s ystem directly estimates the accident severity by comparing the obtained data
with information c oming from previous acc idents s tored in a database. This information is of utmost
importance, for example, to determine the most suitable set of resou rc es in a rescue operation. Since
we want to consider the information obtained just when the accide nt occurs, to estimate its severity
immediately, we are limited b y the data automatically retrievable, omitting other information, e.g.,
about the driver’s degree of attention, drowsiness, etc.
A. Architecture Overview
Figure 1 presents the overview of the vehicular arc hitecture used to develop our system. The
proposed system consists of several compon ents with different func tions. Firstly, vehicle s should
incorporate an On-Board unit (OBU) responsible for: (i) detecting when there has been a pote ntially
dangerous impact for the occupants, (ii) collecting available information coming from sensors in the
vehicle, and (iii) communicating the situation to a Co ntrol Unit (CU) that will accordingly address the
handling of the warning n otification. Next, the notification of the detected accidents is mad e through
a comb ination of b oth V2V an d V2I communications. Finally, the destination of all the c ollected
3
information is the Control Unit; it will handle the warning notification, estimating the severity of the
acciden t, and communicating the incident to the approp riate emergency services.
The OBU definition is crucial for the proposed system. This device mus t be technically and
economically feasible, as its adoption in a wide range of vehicles could become massive in a near
future. In ad dition, this system shou ld be open to future software up dates. Although the design of
the hardware to be included in vehicles initially consisted of special-purpose systems, this trend is
heading towards g eneral-purpose systems be cause of the constant inclusion of new services.
The information exchange betwee n the OBUs and the CU is made through the Internet, either
through other vehicles acting as Internet gateways (via UMTS, for example), or by reaching infras-
tructure units (Road-Side Units, RSU) that provide this service. If the vehicle does n ot get direct
access to the CU on its own, it can generate mes sages to be broadcast by nearby veh ic le s until they
reach one of the aforementioned c ommunication paths. These messages, when disseminated a mong
the vehicles in the area where the a ccident took place, als o serve the purpos e of alerting drivers
traveling to the accident area about the state of the affected vehicle, and its possible interference on
the n ormal traffic flow [6].
Our proposed architecture provides: (i) direct communication between the vehicles involved in the
acciden t, (ii) automatic sending of a data file containing important information ab out the accident
to the Control Unit, and (iii) a preliminary and automatic assessment o f the damage of the vehicle
and its occupants, based on the information coming from the involved vehicle s, and a database of
acciden t reports. According to the reported information and the preliminary accide nt estimation, the
system will alert the required rescue resources to optimize the accident assistance.
B. On-Board Unit structure
The main objective of the proposed OBU lies in obtaining the available information from sensors
inside the vehicle to determine when a dangerous situation occurs, and reporting that situation to the
nearest Control Unit, a s well as to other nearby vehicles that may be affected.
Figure 2 shows the O BU system, which relies on the interaction between senso rs , the data acqui-
sition unit, the processing unit, and wireless interfaces:
• In-vehicle sensors. They are required to detect accidents and provide information about its causes.
Accessing the data from in-vehicle sensors is possible nowadays using the On-Board Diagnostics
(OBD) standard interface [7], whic h serves as the entry point to the vehicle’s internal bus. This
standard is mandatory in Europe and USA since 2001. This encomp asses the majority of the
vehicles of the current automotive park, since the percentage of compatible vehicles will keep
growing as very old vehicles are replaced by new ones.
4
