Lane-level map-matching based on optimization

We propose a method for ego-lane estimation that can robustly determine the currently used lane as required by future lane-precise navigation systems. It employs a lane-level map-matching on a digital road map through least-squares optimization and only requires sensors available in current production vehicles, such as GPS, odometry, visual lane-marking detection and radars. Radar data is used in a RANSAC-based filtering step for lane hypotheses and, together with camera data, in the determination of the reliability of each lane hypothesis. Detailed evaluation in actual traffic in urban scenarios shows very low error rates below 0.2%.