Safety critical event detection - Applying and evaluating different Surrogate Safety Measures in a roundabout traffic scenario

Abstract

Even though surveys have proven roundabouts to be safer than classic intersections, they still are a source of critical traffic encounters especially when entering and leaving them. Traffic participants’ lack of experience on how to drive in this environment and multiple interactions between different types of road users (RU), such as vulnerable road users (VRU) and motorized road users, (MRU) make near-crashes likely to arise. Therefore, VRU-MRU (zebra crossing when the motorist enters or exits) and MRU-MRU (yield situation for entering or leaving the roundabout) scenarios were investigated in a real-world traffic scenario. The collection of critical events from recorded real traffic data is of value for better understanding critical situations and for the development of further automated driving functions (ADF). One of the weak points of evaluating the performance of automated driving functions is that they mostly rely on synthetic simulated traffic scenarios. Feeding this simulated environment with real-traffic based near-crashes will allow testing of ADFs under more reliable circumstances. Within the framework of the EU-funded project L3 Pilot, for 30 days, video and trajectory data had been collected in an urban roundabout in the city of Wolfsburg, Germany. The so called AIM Mobile Traffic Acquisition, camera-equipped systems able to detect, track and classify traffic participants were used for data collection. The output was a space-time virtual representation of these traffic agents that can be used for behavior analyses. Surrogate safety measures (SSM) are well known metrics sensitive to space-temporal closeness between different road agents. Depending on the applied combination of metrics, a risk prediction, the closeness of an already occurred event and/or its criticality can be determined. Different SSMs will be applied in a roundabout setting to identify safety-critical events between motorists as well as motorists and vulnerable road users. The aim is to draw conclusions on the performance of the applied metrics and on the quality of detection of safety-critical encounters. For this purpose, non-normative driving behaviors will be triggered by addressing a risk-related analysis. To do so, different combinations of surrogate safety measures will be applied and the suitability of the metrics investigated and evaluated. The application of SSM on the recorded trajectory data is the key to identifying safety-critical events. The risk analysis will cover the following aspects: (1) Development and testing of different metrics, such as time-to-collision (TTC), TTC2 (also taking the current acceleration of the respective vehicles into account), and-post encroachment-time (PET) among others. (2) An empirical performance analysis of these metrics. Since the SSM rely on several dynamic assumptions, a study is made on how sensitive these metrics are to the risk inherent in the different types of road user encounters. (3) A critical event analysis, where parameters such as frequency of incidence and magnitude of the triggered event will be assessed. Therefore, significant statistical distributions will be created helping to evaluate the critical areas in the roundabout and generating knowledge about the most extreme driving situations. These situations can afterwards be used within the development of ADF, which currently rely on modelled traffic simulation scenarios for their validation. This study will support the testing and development of tools for collection of real traffic based near-crash situations data, which later on could serve as a reference for scenario-modelling within the validation process of ADF.