Handling uncertainties in criticality assessment

Abstract

Reliability is the most important criterion in the design of advanced driver assistance systems. In this context the use of probabilistic activation conditions for e.g. driver warnings or automatic system interventions allows handling of noisy input data and thus reduction of false alarms. In this paper we present a novel computation method for the time metric Time-To-Collision (TTC) in complex traffic situations. Our algorithm is based on a computationally efficient, non Euclidean distance function between extended objects. This function is approximated by a interpolating polynomial and TTC is computed by finding a root of this polynomial. The main contribution of this work is the determination of the collision probability. The calculation of this probability relies on the minimum of the aforementioned distance function and statistical linearization via the Unscented Transformation. The accuracy of the presented method was proven in the simulation using the example of a turning left scenario.