Generation of representative pre-crash scenarios across the full severity range using real-world crash data : towards more accurate virtual assessments of active safety technologies
Publication details: Göteborg : Chalmers University of Technology. Department of Mechanics and Maritime Sciences, 2024Description: 68 sSubject(s): Online resources: Notes: Härtill 2 uppsatser Dissertation note: Licentiatavhandling (sammanfattning) Göteborg : Chalmers tekniska högskola, 2024 Abstract: Virtual safety assessment is now the primary method for evaluating the safety performance of active safety technologies such as Advanced Driver Assistance Systems (ADAS) and Automated Driving Systems (ADS), not the least because there are few alternatives. Generating representative crash scenarios is crucial for the assessment to produce valid results. However, the existing crash scenario generation methods face challenges such as limited and biased in-depth crash data and difficulties in validation. To meet these challenges, this thesis proposed a set of novel methods for generating representative synthetic crashes. This thesis demonstrate the methods for a common crash type, the rear-end crash, in which the front of one vehicle collides with the rear of another. The process of generating synthetic rear-end crash scenarios consists of three main steps: 1) parameterizing the rear-end crashes by modeling the two involved vehicles using naturalistic driving and pre-crash kinematics data, 2) building multivariate distribution models for the parameterized crash data, and 3) generating representative synthetic crash scenarios. Paper A utilized a piecewise linear model to parameterize the lead-vehicle speed profiles in rear-end crashes from two United States datasets. These parameterized speed profiles were then combined and weighted to create a comprehensive dataset representative of lead-vehicle kinematics in rear-end crashes across the full severity range, from physical contact to high severity. Synthetic speed profiles, generated using multivariate distribution models built on the dataset, were then compared with the raw profiles. In Paper B, a following-vehicle behavior model was created by combining two existing driver behavior models. A representative dataset of the initial states (i.e., speeds of both vehicles and the following distance) of rear-end crash scenarios and the minimum accelerations of both vehicles was developed by weighting and combining crash data from various sources.
Härtill 2 uppsatser
Licentiatavhandling (sammanfattning) Göteborg : Chalmers tekniska högskola, 2024
Virtual safety assessment is now the primary method for evaluating the safety performance of active safety technologies such as Advanced Driver Assistance Systems (ADAS) and Automated Driving Systems (ADS), not the least because there are few alternatives. Generating representative crash scenarios is crucial for the assessment to produce valid results. However, the existing crash scenario generation methods face challenges such as limited and biased in-depth crash data and difficulties in validation. To meet these challenges, this thesis proposed a set of novel methods for generating representative synthetic crashes. This thesis demonstrate the methods for a common crash type, the rear-end crash, in which the front of one vehicle collides with the rear of another. The process of generating synthetic rear-end crash scenarios consists of three main steps: 1) parameterizing the rear-end crashes by modeling the two involved vehicles using naturalistic driving and pre-crash kinematics data, 2) building multivariate distribution models for the parameterized crash data, and 3) generating representative synthetic crash scenarios. Paper A utilized a piecewise linear model to parameterize the lead-vehicle speed profiles in rear-end crashes from two United States datasets. These parameterized speed profiles were then combined and weighted to create a comprehensive dataset representative of lead-vehicle kinematics in rear-end crashes across the full severity range, from physical contact to high severity. Synthetic speed profiles, generated using multivariate distribution models built on the dataset, were then compared with the raw profiles. In Paper B, a following-vehicle behavior model was created by combining two existing driver behavior models. A representative dataset of the initial states (i.e., speeds of both vehicles and the following distance) of rear-end crash scenarios and the minimum accelerations of both vehicles was developed by weighting and combining crash data from various sources.