Uncertainty and conceptual model of a camera system in a car crash test scenario

Abstract. Accurately estimating injury severity in crashes relies on understanding vehicle occupant movements. This is simulated using crash test dummies in controlled test cases. Currently, stationary high-speed cameras positioned outside the vehicle track the kinematics of the different dummy parts by following optical markers placed on these dummies. However, onboard high-speed cameras are primarily used for documentation and are not suitable for determining 3d object kinematics with the required accuracy. Furthermore, with the increasing sophistication of modern airbag systems, points inside the vehicle that need to be visible for the stationary cameras may be obscured by the deployment of airbags. To address this limitations, we propose relocating onboard high-speed cameras inside the vehicle and investigating the resulting uncertainties. The dynamic nature of crash events presents challenges for these onboard cameras to accurately self-localize, given the rapid changes occurring within the vehicle. To overcome this challenge, we introduce a novel method for determining the position and orientation of the onboard stereo camera pair at each time point, followed by an analysis of the uncertainties involved. We use Monte Carlo simulations and bootstrapping techniques to estimate the uncertainties associated with point measurements in crash test scenarios. And therefore we can determine the object kinematics and their related uncertainties inside the vehicle using the onboard high-speed cameras instead of the stationary high-speed cameras.