Boris Buschardt

Bayesian Test Design for Reliability Assessments of Safety-Relevant Environment Sensors Considering Dependent Failures

Machine vision based on automotive environment sensors is the enabling technology for advanced driver assistance systems and automated driving. Due to its important role, the reliability of environment sensing is highly safety relevant and has thus to be assessed and demonstrated during the development of the system. The main challenges associated with this task are low target error rates and the stochastic influence of different uncertain environmental conditions on the sensor performance. As a basis for the reliability assessment of environment sensors we introduce comprehensive performance metrics that allow a formal description of the uncertainties in a digital environmental model. Due to the influence of environmental conditions on the sensor performance, these metrics however are not constant but are random variables themselves. This leads to a hierarchical uncertainty structure including higher order uncertainties. To quantify the influence of the environmental conditions on the sensor reliability, we use a Bayesian hierarchical regression model. The utility of this method is demonstrated with a case study in which the influence of temperature on sensor reliability is examined. The results show that the proposed methodology is capable of identifying and quantifying the influence of the temperature on sensor performance. The introduced metrics and the proposed methodology are an important step towards a formalized reliability assessment of automotive environment sensing. In order to predict if the sensor reliability complies with the target error rates, the presented methodology has to be adapted and extended with additional stochastic methods.