Towards a Sensor Failure-Dependent Performance Adaptation Using the Validity Concept

Statically proving the adherence of a system to its safety requirements specified at design-time provokes overcautious systems with limited performance. Contrarily, dynamically assessing the quality of sensor observations at run-time enables adapting a system’s performance accordingly while maintaining its safety. While this could be shown by Brade et al. [2] for a simulated scenario with one-dimensional sensors, we apply the proposed scheme, that is, the Validity Concept to 3D depth information. In this endeavour, we define a failure model covering the failure types Noise, Outlier, and Illumination Artefacts, define functions to estimate their severity at run-time and represent a 3D point cloud’s quality in terms of validity information. Furthermore, we show that calculated Validity Values correlate with sensor failures impairing sensor observations and enable estimating the quality of subsequent applications.