Low-Cost GNSS Aiding by Visual Odometry, Radar and 3D Maps

Autonomous driving is one big driver also for the GNSS R&D community. Nevertheless, GNSS technology still plays a subordinate role in the automotive sector. This comes from the fact that current GNSS receiver technologies are still lacking accuracy and reliability, especially in urban environments. Especially in urban canyons, high buildings are strongly reflecting and shadowing GNSS signals which makes the GNSS solution unreliable or even unavailable. As a result, OEMs do not want to rely on GNSS and put more emphasis on other technologies like cameras, LiDAR, and radar. Regardless, current autonomous car demonstrators from Google, UBER etc. are using GNSS. In order to cope with the limited accuracy in urban situations, GNSS is combined with inertial sensors and vehicle odometry data. The used GNSS/IMU technology, however, will not be affordable for integration with consumer cars. The main goal of the presented paper is to show different approaches how to improve the accuracy and reliability of urban localization for highly automated driving. Therefore, low-cost GNSS was combined with visual odometry, radar and 3D map data in a sophisticated sensor fusion algorithm to reduce the error budget of the single sensors. An investigation about the potential of the different approaches in combination with GNSS was performed. The achieved results were statistically analyzed. Predictive models were developed for analyzing the effect of surrounding as well as driving conditions and predicting the localization performance of the systems for scenarios that were not included in test drives. The interdisciplinary topics of this project were GNSS-based localization, computer vision, remote sensing, geomatics and statistics.