Interactions with vulnerable road users

Deliverable 44.1 of the EU FP7 project UDRIVE Consortium
Jansen, R.; Lotan, T.; Winkelbauer, M.; Bärgman, J.; Kovaceva, J.; Donabauer, M.; Pommer, A.; Musicant, O.; Harel, A.; Wesseling, S.; Christoph, M.; Nes, N. van

Within UDRIVE there has been a specific focus on pedestrians, cyclists and Powered Two Wheelers (PTWs). These groups of road users are particularly vulnerable in traffic because they lack the protective shell that helps preventing serious injury once involved in a collision. In addition, these transport modes have several features that make them more prone to getting involved in a crash, e.g. related to reduced conspicuity and for the two-wheelers the difficulty to remain in balance, either or not in combination with high speeds. This type of factors make that pedestrians, cyclists and PTWs have a high risk of getting fatally or seriously injured in traffic.

Within UDRIVE, a large amount of ‘naturalistic’ data was collected to get more in-depth insight in the interactions of these groups with passenger cars and trucks. The aim was to identify and understand the everyday behavioural patterns in these interactions as well as the circumstances of conflicts or safety critical events in these interactions. The current Deliverable reported on the analyses and results of a number of specific interaction types.


Data were collected by a naturalistic driving approach. In a naturalistic driving study data is collected by equipping people’s own vehicle with various sensors and cameras and unobtrusively registering characteristics of the vehicle, the driver/rider and the environment over longer periods of time and during normal, everyday trips. The analysis of the interactions of car and truck drivers with pedestrians and cyclists was based on data collected from the participating cars and trucks. The analysis of the safety critical events and interactions of PTWs was also based on data from equipped, naturalistic riding PTWs. Data were collected between October 2015 and May 2017.

Starting point for the analyses of the pedestrian and cyclist interactions was the UDRIVE database with data from 186 car drivers in Great Britain, France, Germany, Poland, and the Netherlands, and from 48 truck drivers based in the Netherlands. By April 2017, the database consisted of a total of 42724 hours of car data, and 41397 hours of truck data. The results related to PTWs stem from 47 motorscooter (125CC) riders in Spain, resulting in a database 859 hours of PTW data (Note that these numbers may be slightly different from other UDRIVE deliverables as the dataset was still growing at the time of writing the deliverables.) Depending on the exact research question, the analyses were conducted on a part of the database that fulfilled the selection criteria, e.g. right turning manoeuvres, straight sections, urban areas, et cetera. The next three sections briefly summarize the main findings with respect to pedestrians, cyclists and PTWs as based on the UDRIVE database analyses.

Interactions with cyclists

The analyses of the cyclist data looked at interactions between cyclists and both passenger cars and trucks.

Safety critical events in interaction

First, we investigated which behavioural and situational factors contributed to the occurrence of what was called Safety Critical Events (SCEs) in these interactions, i.e. to real or near-crashes. A near-crash was defined as a situation which was not planned and required an immediate, urgent evasive manoeuvre by at least one of the conflict partners to avoid a crash. The analysis was based on just over 13,200 hours of car data from 125 drivers collected in Germany, Great Britain, France, Poland, and the Netherlands, and on around 6,000 hours of truck data from 41 drivers collected in the Netherlands.

The analysis of the car/truck-cyclist interactions revealed very few SCEs. Overall 11 SCEs were identified: three in interactions with a car, and eight in interaction with a truck. All were near crashes; no actual crashes have been found in the database. All SCEs took place on urban roads with a speed limit of 50 km/h or less. An explanation could be that there are less encounters between cyclists and motorised vehicles on higher speed roads. Given the small number of SCEs only a qualitative analysis was conducted. That indicated that the identified SCEs were caused by a combination of features of the infrastructure (a curve or a too narrow road), features of the manoeuvre (often overtaking), the presence of other traffic, and an error or unexpected behaviour of the cyclist (slowing down). Drivers didn’t seem to make any judgment or performance errors in the observed SCEs. None of the drivers were involved in a secondary task or exceeded the speed limit when they started their evasive manoeuvre and nearly all drivers avoided a collision by further decreasing their speed.

Interactions at intersections and roundabouts

We then zoomed in on a specific type of interaction between vehicle drivers and cyclists, notably interactions on intersections and roundabouts. A first analysis looked at the looking behaviour of car drivers who turned right (left in the UK) passing the path of a (potential) cyclist who wants to go straight through the intersection. This is the typical scenario of a blind-spot crash. The final dataset consisted of 961 intersection manoeuvres by 69 drivers from France, the Netherlands, Poland, and United Kingdom. Furthermore, there were 826 roundabout manoeuvres by 46 drivers from France, the Netherlands, and United Kingdom. Approximately half of the data stem from the United Kingdom, due to it being available early in the project. The results show that on average car drivers actively check the blind spot, i.e. by looking over their shoulder, in around 8% of the cases at intersections and around 4.5% of the cases at roundabouts. Car drivers mostly (between 65 and 95% of the cases) looked in the direction of the road into which they intended to turn, followed by the directions ‘elsewhere’ and ‘sidewalk’. Checking the ‘blind spot’ was done least often. There was a large difference between the investigated countries. On average, at intersections, Dutch car drivers checked their blind spot 6 times more often than drivers in the other three countries (in 27% of the cases), and at roundabouts they did so 21 times more often (in 19% of the cases). The most logical explanation for this difference is that in the Netherlands the prevalence of cyclists and bicycle lanes is higher.

A second analysis of the interactions at intersections and roundabouts focused on the looking behaviour of truck drivers. For this analysis the final dataset consisted of 159 right turn manoeuvres by 10 truck drivers and 209 roundabout manoeuvres by largely the same 10 truck drivers. All of the drivers were Dutch, driving in the Netherlands. On average, truck drivers were observed to check the blind spot in 19% of the cases at intersections and in 27% of the cases at roundabouts. Compared to Dutch car drivers, these Dutch truck drivers checked their blind spot somewhat less often at intersections, and somewhat more often at roundabouts. It should be noted, however, that some of the trucks may have had in-vehicle camera information about the situation in the blind spot, and hence they had no need to turn their head or make large eye movements.

Overtaking manoeuvres

Finally, we had a look at car-cyclist interactions during overtaking manoeuvres. A total of 147 overtaking manoeuvres were analysed. These were manoeuvres by 41 car drivers from France, Germany, Poland and United Kingdom, and concerned rural roads only. It was found that on average overtaking manoeuvres took 9.3s (± 3.5s) and the car speed during overtaking was 61km/h (± 15km/h).

A distinction was made between ‘flying’ overtaking and ‘accelerating’ overtaking. It is called a flying overtaking manoeuvre when the speed of the overtaking vehicle speed remains more or less constant before and during the overtaking. It is called an accelerating overtaking manoeuvre when the overtaking vehicle first stays behind the cyclist and then starts overtaking by increasing its speed. Around 70% of the overtaking manoeuvres was found to be ‘flying’, apart from Poland, where around 50% of the overtaking manoeuvres was ‘flying’.

The main variable of interest in this analysis was the lateral distance between the car and the bicycle, during the actual overtaking manoeuvre. On average the lateral distance was 1.65m (± 0.64m). This is close to the lateral distance of 1.5m that most European countries require by law for overtaking. There were several factors, however, that affected the actual lateral distance. Lateral distances were larger when the speed of the car was higher, when the speed of the cyclist was higher, and when the overtaking vehicle was following another vehicle. Lateral distances were found to be smaller when the cyclist was positioned further away from the edge of the road (towards the centre of the road), when (in case of a flying overtaking manoeuvre) the car driver was a woman, and (in case of accelerative overtaking manoeuvres) when there was an oncoming vehicles.

Interactions with pedestrians

For detecting interactions between cars and pedestrians, the cars were equipped with a Mobileye system. This system provides continuous measures of the distance of the car to ‘objects’ around the car, including pedestrians, calculating, for example, the expected time-to-collision. A detailed analysis of the carpedestrians interactions was based on car data from Great Britain and France. It could be concluded that the real dangerous interactions (real or expected conflicts) were associated with higher car speeds than less dangerous interactions, and required more severe braking. Just over 400 conflicts were identified using a collision warning signal that was switched off for participants, but available to the researchers. The conflicts could be clustered into four subgroups linked to the car’s speed profile.

  1. Conflicts that involved the highest speed group mainly concerned a situation in which the pedestrian (still) was on the pavement.
  2. Conflicts that involved a group of car drivers that had just increased their speed before the conflict occurred; again generally a conflict conflicts in which with a pedestrian was who (still) was on the pavement.
  3. Conflicts in which the high speed drivers probably had noticed the potential conflict well in advance, and had reduced speed to avoid a collision.
  4. Conflicts in which the car driver had not reduced speed until very late, seemingly because he had not at all noticed the pedestrian. This group of potential conflicts contained the highest percentage
  5. of real conflicts (SCEs).

As indicated, the current study used the Mobileye system as a means to identify interactions with pedestrians. Originally, however, this system is meant to be an in-vehicle system that warns drivers when they approach a pedestrian. Based on the UDRIVE data it was investigated whether this system, if used as a warning device, would indeed be able to provide the correct and relevant information to the driver. It was concluded that in some cases an early alert as provided by Mobileye may be potentially beneficial for preventing a conflict to turn into a real collision. Analysis of the videos showed that the large majority of (expected) conflicts as identified by the system were indeed (potential) conflicts. Hence, the system is good and relevant for detecting potential conflicts with pedestrians. In around three quarters of these situations, the driver him/herself had spotted the pedestrian in time. In the still substantial share of remaining situations, a warning system could have been of help. A warning system can be expected less useful in conditions with relativly many pedestrians. In those cases car drivers appeared to be already more alert to pedestrians' presence and potential conflicts.

Interactions with PTWs

Where information about pedestrians and cyclists was inferred from the data collected by the instrumented cars and trucks, the information about the powered two-wheelers (PTWs) also comes from instrumenting the PTWs themselves, i.e. from Naturalistic Riding. The work on PTWs looked at the possibilities and challenges of identifying conflicts or safety critical events. Furthermore, it looked at characteristics of everyday riding with a special focus on speed choice and acceleration at urban intersections, and on the distance (time headway) between cars and PTWs on straight road sections.

The identification of safety critical events

Obviously, PTWs have their own very specific dynamics, posing specific requirements to the data collection equipment and to the interpretation of the collected data. Some of the previous attempts with Naturalistic Riding showed that one of the challenges is the identification of safety critical events (SCEs). In our study SCEs were identified by looking at a set of kinematics-related variables (including longitudinal acceleration, lateral acceleration, vertical acceleration, rotation speed) and identifying the extremes or outliers: the high-g events. For these events, the video material was studied to assess if there had actually been an SCE and in case it had, to identify the circumstances related to rider, other traffic and infrastructure.

Analyses were based on 497 hours of data (equalling 13.654 kilometres driven) from 39 riders in Spain. A total of almost 1,300 potentially relevant events were identified based on the motion-related variables. Because only around 70% of the video registrations were usable, around 500 events could be checked based on video registration. The vast majority of the identified events appeared to be related to a non-safety relevant manoeuvre, such as a speed bump, a tight curve, starting from or braking to a stand-still, entering or leaving a parking lot, etc. In other words there were a large amount of ‘false alarms’. Only two safety relevant events were identified based on these high-g events. One was based on an extreme longitudinal acceleration (harsh braking) in a one directional dual lane situation where the view off a pedestrian who started to cross at a zebra crossing was blocked by vehicles in the other lane. The other was based on extreme lateral acceleration (swerving) due to a passenger car entering from a side road into the path of the motor rider. Obviously, based on this approach it is unknown how many SCEs were missed. Situations in which it is the other road user who takes evasive actions rather than the motor rider who might not even have perceived the potential hazard, will never be identified based on g-forces from the motor cycle.

Characteristics of everyday riding behaviour

This analysis focused on speed choice and acceleration by PTW riders in four common urban intersection scenarios: free flow followed by a right turn, free flow followed by a left turn, full stop followed by a left turn, and full stop followed by a right turn. The analysis was based on 7350 manoeuvres by 32 riders, where each rider featured at least 10 manoeuvres in at least one of the above scenarios.

There are two main findings in this study. First, significant differences have been found between the scenarios. Pair-wise comparisons showed that most scenarios were significantly different from each other on all measures, these being speed at the manoeuvre onset, speed at the manoeuvre offset, average speed, maximum speed, minimum speed, acceleration at the manoeuvre onset, average positive and negative acceleration, and maximum positive and negative acceleration.

The second main finding concerns a comparison between riders. Across riders significant differences have been found in speed choice and acceleration during manoeuvres, as well as in the time window surrounding full stops prior to the manoeuvres. Furthermore, riders appear to use a constant deceleration in the five seconds preceeding a full stop, but the magnitude of this deceleration varies across riders. These findings suggest that riders have different preferences (i.e., riding styles) regarding speed choice and acceleration.

If such preferences indeed exist, they may inform the development of intelligent warning systems on what is ‘normal’ and ‘abnormal’ riding behaviour. Furthermore, the existence of preferences warrants further research on whether groups of riders share similar preferences. This could be done with a bottom-up, or data-driven, approach (e.g., cluster analysis), or through a top-down approach (e.g., with behavioural questionnaires).

Time headway between cars and PTWs

This analysis focused on the time headway, i.e. the following distance expressed in seconds, on straight sections of roads between cars and PTWs in comparison to the time headway between two cars and between cars and trucks. For this analysis the starting point was the car. Data came from 140 car drivers from France, Germany, Netherlands, Poland and the United Kingdom who together had driven almost 650,000 km and waswhich were searched to identify relevant interactions. Final analyses included over one hundred million situations where the car was behind another car, over 6 million situations where the car was behind a truck and almost 370,000 situations where the car was behind a PTW. Different road types with different speed profiles were included in the analysis.

Overall, the time headways for following another car, a truck or a PTW were very similar. At lower driving speeds (< 50km/h) the average time headways were around 1.7s, at medium speeds (60 - 80km/h) the average time headways varied somewhat between 1.4 and 1.6s. At speed over 80km/h the time headway in car-car situations remained around 1.4s, but the time headway in car-truck situations tended to increase again to around 1.7s. Whereas the general picture showed very similar time headways for the different vehicle combinations there are two exceptions worth mentioning: cars followed trucks slightly closer than they followed other cars and PTWs, and at medium speed cars followed PTWs at a slightly longer distance than cars or trucks. There were hardly any differences between the five countries in the choice of time headway. We just saw that the German car drivers seemed to keep somewhat more distance behind trucks at medium speed, and the French car drivers seemed to keep somewhat less distance to other cars. Distances to PTWs were very comparable between countries. All together the data did not show that car drivers tend to follow PTWs closer than cars or trucks. There was even an indication that car drivers followed at some larger distance.


Overall it can be concluded that Naturalistic Driving is a very interesting method to collect in-depth and valid insights in road user behaviour. Compared to previous large Naturalistic Driving studies (e.g., 100 Cars, SHRP2), UDRIVE has a unique focus by including interactions with vulnerable road users, both from the perspective of car and truck drivers, as well as from the perspective of powered two-wheelers. Rather than focusing exclusively on crashes, the interactions with vulnerable road users have been studied at varying levels of criticality, ranging from Safety Critical Events and blind spot checks to overtaking manoeuvres and everyday riding. The findings have given rise to recommendations on vehicle safety, for awareness campaigns and training, and on the design of road infrastructure. It is our hope that the recommendations, once implemented, will improve the safety of vulnerable road users, and in this way contribute to the EU target of halving the number of road deaths by 2020.

Gepubliceerd door
European Commission, Brussels


Dit is een publicatie van SWOV, of waar SWOV een bijdrage aan heeft geleverd.