Analyzing real-world data to promote development of active safety systems that reduce car-to-vulnerable road user accidents

University dissertation from Chalmers University of Technology

Abstract: The overall objective of the thesis is to explore various types of real-world road traffic data and to assess the extent to which they can inform the design of active safety systems that aim to prevent car-to-vulnerable road user (VRU) accidents. A combined analysis of in-depth and police reported accident data provided information on driver behavior and contextual variables, which is valuable for the development of active safety systems. An analysis of the in-depth data also revealed information about VRU behavior that is relevant for these systems. A key finding from these analyses is that the car drivers commonly did not see the VRUs due to visual obstructions in the traffic environment, misinterpretation of the traffic situation, and/or an inadequate plan of action. The VRUs, on the other hand, saw the cars but they still misunderstood the situation, made an inadequate plan of action, or both. These findings indicate that active safety systems should help drivers to notice the VRUs in time, while the VRUs would benefit from systems helping them to correctly understand the traffic situation. The findings also suggest a need for a variety of cooperative active safety systems, risk assessment algorithms able to predict the intentions of road users to cross the road, and human-machine interfaces capable of directing road users’ attention towards the most critical event. Similar findings were obtained when driver behavior and contextual variables were investigated using video-recordings of car-to-pedestrian incidents. However, these data enabled more detailed analysis of driver attention allocation as well as driver interaction with the vehicle, other road users, and the traffic environment. Finally, an analysis of data on pedestrian behavior and car dynamics from normal interactions in traffic showed that a statistical model, based on car speed and its distance to the point of potential collision and on pedestrian distance to the road, speed and head orientation, could be used to determine the likelihood of a pedestrian entering the road. This can then be combined with commonly used deterministic approaches to estimate when a warning or other action by an active safety system should be initiated. To conclude, each of the four data sources explored here has its own advantages and disadvantages; information combined from analysis of these sources provides an improved understanding of the traffic situations involving VRUs, which is crucial in the development of future active safety systems.