Abstract

The main objective of the project PROLOGUE (PROmoting real Life Observations for Gaining Understanding of road user behaviour in Europe) is to explore the feasibility and usefulness of a large-scale European naturalistic driving observation study. The work described in this deliverable focused on the feasibility of using and combining two observation methods, site-based observation and in-vehicle observation, for naturalistic driving analysis. In this study the strengths and weaknesses of each method were analyzed and the differences between measures that were calculated by both methods were compared to consider the combined results. The two different observation methods studied the interaction between car drivers turning right and cyclists going straight at an intersection, which was equipped with cameras. For the site-based study one camera was used that had a bird’s eye view on the intersection. The video data were processed manually and automatically. The results of the two ways of analysis were compared. The instrumented cars were equipped with a data acquisition system (DAS) and sensors including a GPS logger and a camera. The results of the site-based observations showed that the manual processing seemed to be more accurate than the automatic processing. However, the processing time was much larger for manual analysis. Having a top view camera or more cameras with different viewpoints could improve the accuracy of automatic processing by providing the ability to detect the sizes of the road users and therefore the road user type. The in-vehicle measurement showed that drivers show different driving behaviour when making the right turn manoeuvre from a halted situation compared to a non-halted situation. This implies that drivers adapt their behaviour and compensate for the more dangerous situation by driving slower and looking more often. The results of the speed measure comparison between the site-based and in-vehicle study showed that the average speeds for both studies were similar. However, the speed profile that was measured in the in-vehicle study was less smooth than the speed pattern that was derived in the site-based study. This is explained by the noisy GPS signal on which the in-vehicle speeds are based. The value of combining the two observation methods is twofold. At first it offers the opportunity to enrich the information from one study with complementary observations from the other study. Secondly, it offers the opportunity to validate measures from the individual studies. This study showed that each type of study has its unique values. From in-vehicle data it is possible to look in detail at the driving behaviour of the participants over time and in different natural situations. The video data allows looking over the shoulder of the driver and observer where drivers are looking or what they are paying attention to. By analyzing their glancing behaviour, we know when and how often drivers look at other vulnerable road users to estimate their position and speed and to be able to anticipate. For a future large scale naturalistic driving study it could be valuable to add a number of sites with site-based observation. Two main areas are identified that would benefit from these complementary observations. One research area is when studying the interaction between drivers and vulnerable road users. Another research area that would benefit from the complementary observations is research questions related to infrastructure – as these are by nature related to a specific site like an intersection, a lane merge or a roundabout. The site-based observation allows to observe the traffic flow, the in-vehicle observation could add (for the overlapping cases) information on looking behaviour or other in-vehicle behaviours as well as additional vehicle data like accelerations.

Abstract

There is a pressing need for road authorities to take a proactive role in the deployment of automated vehicles on the existing road network. This requires a comprehensive understanding of the road infrastructure requirements that would lead to safe operation of automated vehicles. In this context, a field test with Lane Departure Warning and Lane Keeping Systems-enabled vehicles was conducted in the province of North Holland, The Netherlands. The performance of these automated systems was evaluated using performance indicators such as Mean Lateral Position and Standard Deviation of Lane Position. In this study, the Systems Theoretic Accident Modelling and Processes (STAMP) model was adopted to understand the relationships between the various components of the “Road System”, which in this study include the road authority, the automated vehicle system, elements of the road infrastructure, and weather conditions. Empirical data from the experiment is used to estimate the relationships between the different components, followed by the assessment of their impact on the performance of the automated vehicles. It was found that visibility conditions have a significant effect on detection performance, which worsens in rainy conditions especially under streetlights. It has been also observed that there is a significant difference in Lane Position between Left Curves and Straight sections, and between lane widths less than 250 cms and those that have larger widths. These findings are combined with the results from the STAMP analysis to formulate a set of road infrastructure requirements that would lead to safe performance of Lane Assistance Systems.