Detecting and Evaluating Risky Behaviors in Overtaking Scenarios for Safer Driving

Abstract

While cycling has become a popular way of modern transportation, concerns about cyclist safety remain significant. One typical and potentially risky scenario arises when motor vehicles pass (including overtake) cyclists, which may lead to close passing incidents and endanger the vulnerable road users. Previous works intensively investigated a typical scenario where an ego vehicle is overtaking a cyclist on a two-lane rural road with possible oncoming vehicles present on the adjacent lane. Experiments have been conducted to study the characteristics in vehicle-cyclist interaction during overtaking. Many studies segment the overtaking period into phases to better model the maneuver. There are also traffic laws that requires a minimum lateral distance when a motor vehicle passes cyclists. In this thesis, we used real-world data from Volvo Cars Corporation to detect and evaluate risky passing and overtaking behavior. We used and further developed a tool that extracts scenarios when ego vehicles pass cyclist. Then output information derived from signals in these scenarios. With that information, we built a rulebased model and applied it to different phases of the overtaking process. The model used four safety metrics: Variable-Adjusted Minimum Passing Distance, Perceived Risk Score, Minimum Distance Returning, and the Lateral-Time Risk Index. We conducted statistical analysis on the data and the model output. We also verified some of the results by manually reviewing the scenario on a visualization system to evaluate the correctness of our model. Furthermore, we discussed some special or difficult cases, such as overtaking a cyclist group, sensor unreliability at night, overtaking maneuvers happening on curved roads. We also pointed out the possible limitations in the data source, sensor fusion process and other factors. Lastly, we showed our perspective on the future works that can be based on our thesis work. The work presented in this thesis can help better understand how drivers overtake cyclists, and it may be useful for providing post-event feedback to enhance driving safety, or provide insights to user-based insurance systems.