Efficient Localization on Highways Employing Public HD Maps and Series-Production Sensors

Compared to standard navigation maps, HD maps contain precise additional information for automated vehicles. To exploit this information, a lane-level accurate localization estimate within the HD map needs to be available. Here, the computation and memory overhead of the localization algorithm needs to be as small as possible to enable a reasonable usage of HD map information. Since the accuracy of GNSS measurements is insufficient for an accurate localization at lane-level, HD maps commonly contain geo-referenced landmarks like lane markings and boundaries that can be detected by series-production camera sensors, especially on highways. These measurements are fused with the motion measurements provided by the ABS and ESP sensors of the vehicle within a Bayesian filtering framework. Here, the most popular localization algorithm coming from the field of robotics is the particle filter (PF). However, for a robust estimation, a sufficiently high number of particles is required which is in contradiction to the objective to minimize the computational and memory overhead of the localization algorithm. Hence, in this contribution a memory and run-time efficient Extended Kalman-Filter (EKF) based solution is proposed employing measurements from series-production vehicle sensors and a public HD map from two highways near Düsseldorf. Experimental results achieved with the EKF are comparable with the localization accuracy provided by the PF while run-time and memory consumption are considerably reduced.