[Back]

K Schwieger, A. Hula, P Saleh, H. Ecker, M. Neumann:
"Avoiding Motorcycle Accidents by Motorcycle Risk Mapping";
Talk: Online-IFZ-Konferenz 13th International Motorcycle Conference, Deutschland; 2020-09-01; in: "Web-Proceedings of 13th International Motorcycle Conference", (2020), 8 pages.

The risk of suffering a motorcycle accident is still difficult to quantify given the different types of individual riding styles of motorcyclists. Every year, nearly 100 motorcycle accidents occur on Austrian roads [1], but the rare occurrence of accidents at the same spot makes it difficult to locate risky spots. To tackle these challenges, test riders were gathering data while driving on popular motorcycle routes with a motorcycle equipped with the latest sensor technology in order to collect individual driving dynamics data. The test vehicle used (MoProVe, Motorcycle Probe Vehicle) was a modern KTM motorcycle that is equipped with cameras, geo-position antennas (GPS, GLONASS), inertial measurement units (IMUs) and with access to the motorcycle's internal data acquisition (CAN bus).

In order to achieve the identification of risky spots from driving dynamics, the project viaMotorrad (funded by the Austrian Road Safety Fund - VSF) [2] started in 2015 with the aim of developing tools and methods for improving motorcycle safety in the national road network. The test vehicle MoProVe was developed by the Vienna University of Technology and the AIT-Austrian Institute of Technology in order to obtain the necessary driving dynamics data. It was the focus of the project to enable a preventive / predictive approach to motorcycle safety based on the dynamic driving data instead of the conventional, reactive approach.

This work presents a central part of the final project results of viaMotorrad: A method for identifying accident hot spots with risky driving dynamics parameters through machine learning. The developed approach uses a combination of unsupervised and supervised learning methods in machine learning to distinguish the dynamics at known accident sites from the most common driving dynamics of every single driver in a pool of 6 drivers. The individual risk estimates are then combined into a common risk estimate for the pool of test drivers in the experiment. Using the joint estimate, maps of risk spots for 6 popular motorcycle routes in the Austrian Alps were derived. The risk spots on these maps were further divided into three groups depending on how high the individual risk warnings were in their vicinity. Therefore, these maps may include a priority checklist for implementing security measures on popular routes.

https://publik.tuwien.ac.at/files/publik_294616.pdf