Evidential Multisensor Fusion and Erroneous Management of Lanes for Autonomous Driving

Abstract

Lane information is essential for safe autonomous driving. In this article, we present a multisensor fusion framework for ego and adjacent lanes with a novel fusion quality measure and dynamic lane mode strategies for erroneous management. The framework fuses road marking lines based on Dempster-Shafer theory and tracks lanes with a particle filter. Then, a quality measure for each line is computed, integrating sensor coherence, availability as well as temporal continuity. This quality is essential to deploy different lane management strategies in order to avoid integrating erroneous data. The proposed framework was evaluated in a lateral control architecture with autonomous driving on open roads and proved its robustness and availability.