Improving Driver Safety through the Identification, Prediction, and Warning of Traffic Accident Hotspots
Open access
Author
Date
2018Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
Across the globe, injuries sustained from traffic accidents are the eighth leading cause of mortality, and with the number of annual deaths steadily rising to over 1.25 million, now account for 2.5 % of total worldwide fatalities. This growing issue is not limited to the low and middle-income regions of the world, as the frequency and severity of traffic accidents has also been increasing in developed countries over the last decades. For example, between 2014 and 2015 the amount of traffic fatalities in the United States rose sharply by 7.2 %. Through analysing the patterns and locations of traffic accidents, road authorities can identify dangerous sections of the road network and prioritise these locations for infrastructure improvement, helping to prevent these tragic events from occurring. However, traffic accident analysis is traditionally based on historic crash data and is restrictive in many ways, typically suffering from issues including small sample sizes, underreporting of traffic accidents, and data scarcity. Furthermore, in the limited number of countries where it is available, historical accident data is often only provided on a deferred basis and analyses can be severely out-of-date.
Naturalistic driving data, available from the advanced sensors and technology em- bedded in connected, semi-, and fully-autonomous vehicles, potentially offers both road safety researchers and practitioners a new and dynamic source of variables for analysis. The technology in these vehicles can be leveraged to detect accidents and ‘near miss incidents’, or critical driving events, such as heavy braking and evasive manoeuvres, and reliably predict locations with a high likelihood of traffic accidents. Both researchers and industry players alike see the promise of this data to combat the existing challenges of accident analyses. For example, real-time assessment of the locations of these events could aid road authorities in monitoring existing accident hotspots, as well as identifying new and developing areas of high accident exposure, offering various possibilities to intervene before incidents occur there. Yet, despite the great potential in identifying the locations of traffic accident hotspots with insights from these vehicles, to date, there is limited empirical evidence on whether the perilousness of locations can be accurately predicted through naturalistic driving data.
Furthermore, with these insights and the rise of increasingly connected and intelligent vehicles, as well as the emergence of smartphone turn-by-turn navigation applications, various safety-focused innovations become a possibility, such as providing safe-routing services and in-vehicle warnings of potential accident hotspots. Whereas safe-routing will attempt to avoid an accident hotspot entirely, encounter- ing these dangerous locations will always remain a possibility. Consequently, identifying ways of effectively reducing the frequency and severity of traffic accidents at these known locations remains of the utmost importance. Latest studies provide promising evidence that in-vehicle warning systems can have significant positive effects on driving behaviour and collision avoidance, and while the potential of these systems to improve driver safety are undisputed, the vast majority of studies have focused on simulation setups or controlled field experiments. Moreover, the benefit of real world location analytics on accident hotspots as a data source for in-vehicle warnings has widely not been investigated.
In order to address the aforementioned research gaps, a comprehensive intervention system was developed as part of the research at hand and deployed in a realistic field setting. This system both collected naturalistic driving data and provided warnings to drivers based on location analytics applied to a national historical accident dataset, composed of over 266 000 accidents. As such, this thesis depicts the design and field evaluation of an in-vehicle system, that for the first time bridges the gap between real world location analytics, accident hotspot warnings, and a naturalistic driving setting. The presented system was deployed in an 18 week nationwide field study of 72 professional drivers, covering over 690 000 km, and collected high frequency sensor data from the CAN Bus of each of the vehicles.
Ultimately, by going beyond existing research and exploring driver behaviour in a naturalistic driving setting, this thesis demonstrates that in-vehicle warnings of accident hotspots had a significant improvement on driver safety over time. First evidence is additionally provided that an individual’s personality plays a key role in the effectiveness of such in-vehicle warning systems. However, in contrast to the promising results of existing lab experiments, an immediate effect on driver behaviour was not observed, further highlighting the importance of conducting experimental research in a realistic field setting.
This thesis additionally identifies the potential of driving data to reliably predict the locations of accident hotspots, assessed through a nationwide spatial regression to determine Crash Frequency across the majority of the Swiss road network. The results demonstrate a proportional relationship between Crash Frequency, and heavy braking events and trip frequency measurements from the field study fleet, along with additional explanatory variables for urban and highway locations. These insights provide initial indications that companies, organisations, and other players in the automotive industry with access to a fleet of connected, semi-, or fully- autonomous vehicles can determine existing and newly arising locations of high accident probability. Such a data-powered approach to road safety both enables the possibility for road authorities to intervene before traffic accidents occur at emerging dangerous locations, and empowers new safety-focused automotive services, such as the in-vehicle warnings that have been shown in this work to encourage safer driving through accident hotspots. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000306251Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichSubject
Road Traffic Accident Analysis; Driver Safety and In-Vehicle Warnings; Potential of Driving Data; The Impact of Accident Hotspot Warnings on Driver Behaviour; Spatial Prediction of Traffic Accidents with Heavy Braking EventsOrganisational unit
03681 - Fleisch, Elgar / Fleisch, Elgar
More
Show all metadata
ETH Bibliography
yes
Altmetrics