|
Citation
|
Thomas L, Jörg M, Martin V, Patrick L, Lennart V N, Syn S. 27th International Technical Conference on the Enhanced Safety of Vehicles (ESV); April 3-6, 2023; Abstract #: 23-0210, pp. 19p. Washington, DC USA: US National Highway Traffic Safety Administration, 2023 open access. |
|
Abstract
|
27th International Technical Conference on the Enhanced Safety of Vehicles (ESV): Enhanced and Equitable Vehicle Safety for All: Toward the Next 50 Years
https://www-esv.nhtsa.dot.gov/Proceedings/27/27ESV-000210.pdf
Urban traffic is characterized by limited traffic areas, varying traffic flows and the occurrence of different types of road users. To further advance automated mobility, the severity of injuries sustained by vulnerable road users (VRUs) in unavoidable accidents must be minimized. The project "ATTENTION", supported by the German Federal Ministry for Economic Affairs and Climate Action, was set up to tackle this issue by developing a method for the real-time prediction of VRU injury risk using artificial intelligence (AI). The present study represents the first step in the ATTENTION project and evaluates behavioral aspects of VRUs in real-life car crash scenarios. Firstly, a comprehensive, hand labeled database of video documented VRU crashes from South Korean dashcams was set up. Secondly, the data was analyzed to determine relevant characteristics like pedestrian pre-crash movement and behavior. Afterwards a comparison against the German in-depth accident study database was performed. Finally, relevant scenarios were extracted, and AI-based preprocessing was applied. Body-shapeestimation methods were used to extract pedestrian poses and kinematics for further statistical processing. In 9,724 video documented crashes, 369 frontal primary collision against VRUs were deemed usable. The analysis reveals that every 4th crash in this sample is potentially not avoidable due to physical limitations. The VRU recognized the car before impact in every 2nd crash, possibly performing evasive actions prior to first impact. Comparisons revealed that 31,000 similar car-VRU crashes were documented in the German In-Depth Accident Study (GIDAS) database. The estimation of plausible shapes and kinematics was possible in 37 of 319 pedestrian cases (12%), while 10 of 50 videos (20%) involving cyclist could be processed. Distinct pre-crash poses and kinematics were objectively identified and were shown to be different from standard gait-cycle kinematics. The VRU shapes and poses were used to define average pre-crash body shape appearances and hull-spaces for use in future human body model simulations. The results of this study show that a VRU pre-crash behavior can be objectively determined from low-quality infield video data using AI-driven methods and that it differs from regular human motion patterns. Furthermore, it shows that this video data can be used to setup a position and movement database. Both lay the foundation to estimate an injury risk index of VRUs in the later stages of the ATTENTION project.
Language: en |