Publication

The UDRIVE dataset and key analysis results

Deliverable 41.1 of the EU FP7 project UDRIVE Consortium

Author(s)

Bärgman, J.; Nes, N. van; Christoph, M.; Jansen, R.; Heijne, V.; Carsten, O.; Dotzauer, M.; Utech, F.; Svanberg, E.; Pereira Cocron, M.; Forcolin, F.; Kovaceva, J.; Guyonvarch, L.; Hibberd, D.; Lotan, T.; Winkelbauer, M.; Sagberg, F.; ... Fox, C

Year

2017

UDrive is a large European naturalistic driving study, sponsored by the European Commission (FP7). Nineteen partners across Europe have come together and, along with stakeholders, defined research questions, developed data acquisition, collected and managed data, and finally, performed a first analysis on the UDrive dataset with respect to driver/rider behaviour related to traffic safety and the environment (ecodriving).

This document presents key results of the UDrive analysis performed in UDrive Sub-project 4: Data analysis. It also describes the UDrive dataset and, in brief, how we got here.

The UDrive dataset consists of 38 157 hours of analysable passenger car data from 192 drivers (collected in five countries: Germany, France, the Netherlands, Poland, and the United Kingdom), 14 503 hours of analysable truck data from 46 drivers (collected in the Netherlands), and 497 hours of analysable powered two-wheeler (scooter) data from 39 drivers (collected in Spain). The descriptive statistics of the data in this deliverable is the status of the dataset (analysable data in the UDrive database) as of June 27th, 2017.

The analysis in UDrive was facilitated by tools such as the quality assurance procedures and data tracking (e.g, the on-line data monitoring tool), the SALSA dataprocessing tool, the UDrive annotation codebook, high-quality manual annotation of video, and a team of highly skilled researchers. The analysis itself is described in short in this report, while details are presented in separate UDrive deliverables. This document summarizes the key results of UDrive, so only a few typical examples are given in this executive summary. For example, in the safety-related analysis of risky and everyday driving, results show how car drivers’ speeding varies across the day, and how males and females speed differently. Note that “light speeding” is defined as exceeding the speed limit by 11%-15%, and “severe speeding” as exceeding the speed limit by 16-20%. (“Extreme speeding”, exceeding the speed limit by >21%, did not occur in this sample.)

In the analysis of distraction and inattention, the focus was on when, where, and how drivers engage in secondary tasks while driving. It is shown to what extent drivers in four different European countries perform secondary tasks (any of the tasks annotated in UDrive: mobile phone use (further broken down into individual actions in other analyses), electronic device use, eating/drinking, smoking, reading/writing, personal grooming, talking/singing, and other) across 653 trips (approximately 167 hours) driving by 87 passenger car drivers. An obvious observation is the large difference by country in secondary task performance while driving. Unfortunately the data from The Netherlands were not annotated in time to take into account in this analysis.

UDrive is unique in that one of its foci is on the safety of vulnerable road users (i.e., the interactions between cars and pedestrians/bicyclists as well as aspects of powered two-wheel driving). An example is how car drivers’ glance behavior (blind spot checks) when turning right (UK left) in intersections, and roundabouts differs across four European countries. A startling insight in UDrive was not only the very low proportion of blind spot checks, but also the large variablility between countries.

Results were obtained from the analysis of just over 400 Pedestrian Collision Warnings (PCWs; from the MobilEye smart camera data) in car data from Great Britain and France. Speed choice and speed management are key factors in all road conflicts—particularly in conflicts involving vulnerable road users, such as pedestrians. Therefore, speed is the most natural choice for clustering PCWs. The PCWs were clustered into four conflict categories:

  • A. Conflicts that involved the highest-speed group, mainly comprising situations in which the pedestrian was on the pavement.
  • B. Conflicts that involved a group of car drivers who increased their speed just before the conflict occurred; again, mainly situations in which the pedestrian was on the pavement.
  • C. Conflicts in which the high-speed drivers had probably noticed the potential conflict well in advance, and had reduced speed to avoid a collision with the pedestrian.
  • D. Conflicts in which the car drivers had not reduced speed until very late, seemingly because they had not noticed the pedestrian. This group of potential conflicts contained the highest percentage of real conflicts (safety-critical events; SCEs).

These four clusters provide clear and distinct speed choice behaviours relating to the occurrence of PCWs. The most interesting is the cluster in which the drivers did not reduce their speeds until the actual onset of the conflict, meaning that drivers were not aware of the conflict until it actually occurred. This cluster has the highest percent of safety-critical events (i.e., 22 of the 67 identified safety-critical events) and the lowest proportion of vulnerable road user (VRU) facilities (e.g., sidewalks etc.).

Finally, eco-driving was a key area of analysis in UDrive. An example of one such analysis are the eco-driving scores calculated in UDrive as the average of the residual values of:

  • Braking energy at 50-60km/h
  • Engine speed when shifting from second to third gear
  • Most frequent (peak) velocity at speed limits between 95 and 120km/h
  • Width of the peak around the most frequent velocity at speed limits between 95 and 120km/h
  • Weighted mean of the absolute acceleration at speed limits between 95 and 120km/h

The rationale behind this eco-scoring is that braking energy is, for most drivers, the main energy consumer at low velocities (larger than rolling resistance or air drag). The difference in lost braking energy between the best and worst drivers is in the order of 120%, resulting in a difference in energy consumption of up to 10%. Further, engine losses are not negligible for passenger cars. Idling in urban areas occurs 15% of the time, with a range of 0-50% in the UDrive dataset. In addition, some drivers shift gear much earlier than others, even in the same type of vehicle. The estimated difference in fuel consumption due to different engine speeds can be as much as 20-25%.

The average of the residual percentage values of the variables described above gives an eco-score—negative for better-than-average and positive for worse-than-average eco-drivers. Since it is expected that a correction for driving circumstances has a large influence on driving behaviour, a selection is made on free-flow circumstances (based on headway), excluding trajectories with bends and intersections. The results show a wide spread in eco-score. Further analysis identified individual aspects of eco-driving and driving style.

In summary, in UDrive a large variety of analyses was performed on the UDrive naturalistic driving data (NDD). Although the efforts and results have been significant and already impact safety measure design and development, the UDrive project has only scratched the surface of the analysis potential, for both safety and eco-driving. Similar to the US government agencies funding the large US naturalistic driving study SHRP2, which expect the SHRP2 data to be ”…useful to transportation safety researchers and others for at least 20 years.” (SHRP2, 2010, p. 1), the UDrive partners believe the UDrive data will be a valuable resource facilitating research, traffic safety, and eco-improvements for many years to come.

Print this page
report

Pages

208

Publisher

European Commission, Brussels