Simulation-based Sensitivity Analyses of Visual Localization in Challenging Environments

Abstract

Digital twins of sensor systems enable in-depth analysis of visual localization methods based on synthetic video clones. This is necessary to develop and prepare them for reliable operation in challenging and hazardous real-world environments. Exemplary for a visual-inertial navigation system, this poster presents experiments from Monte-Carlo-based multiparameter sensitivity analyses to assess significant error sources from different environmental, system design, sensor property and calibration error parameters. Three different synthetic video clones are used that closely replicate datasets from corridor, volcanic coast and fumarole environments, and contain the dynamic elements person, water, and smoke.