A real-time multisensor fusion verification framework for advanced driver assistance systems

Abstract This paper presents a novel approach for the verification of multisensor data fusion algorithms in complex automotive sensor networks. Multisensor fusion plays a central role in enhancing the interpretation of traffic situations, facilitating inferences and decision making. It has therefore been instrumental in the ongoing innovation of Advanced Driver Assistance Systems (ADAS) which paves the way to autonomous driving. We introduce a real-time framework which can benchmark the performance of the fusion algorithms at the electronic system level using a Hardware-in-the-Loop (HiL) co-simulation. The presented research provides a quantitative approach for a trade-off between physical realism and computational efforts of the real-time synthetic simulation. The proposed framework illustrates a generic architecture of ADAS sensor error injection for robustness testing of the System under Test (SuT). We construct a lemniscate model for errors to find multivariate outliers with the Mahalanobis distance. A critical driving scenario considering road users in urban traffic describes the dynamic behaviour testability of the fusion algorithms. The industry-proven framework facilitates a functional verification of multisensor-fusion-based object detection precisely and more efficiently on the target electronic control unit (ECU) in the laboratory.