An Evidential RANSAC Algorithm and Its Application to GNSS Positioning

Salim Zair, Sylvie Le Hégarat-Mascle, Emmanuel Seignez

Abstract: The RANSAC (random sampling consensus) approach was proposed for robust estimation in presence of outliers, that are detected as inconsistent with the solution. In this paper, we adapt its principle to derive an algorithm detecting inconsistent sources based on their modelling in evidential framework. We compare two (in)consistency criteria: the classic empty set mass and Pichon’s consistency measure that was recently proposed. The proposed approach is applied to positioning from Global Navigation Satellite Systems (GNSS), specifically in constrained environments, i.e. in the presence of Non Line Of Sight and multipath receptions. Results are compared with former approaches either in belief functions framework or using interval analysis, stating the interest of the proposed algorithm.