The highly accurate localization of the host vehicle is a major challenge in the subject area of autonomous driving. Although modern DGPS systems provide sufficient accuracy, they are not relevant for series production cars due to two reasons: They are firstly very expensive and secondly not permanently available, because they require the visibility of a sufficient number of navigation satellites and an additional correction data service.
A possible alternative is the landmark-based localization. Here distinctive features of the vehicle surroundings are detected by different sensors (eg. laser, camera, radar) and then compared with a previously learned featured stored in a map. The searched pose of the vehicle can finally be estimated using a Monte Carlo localization implemented by a particle filter.
The aim of the research project "highly accurate localization" is to identify robust features and novel localization approaches. Particularly challenging are hereby procedures that address the SLAM problem (simultaneous localization and mapping). The localization of the subject vehicle and the mapping of the vehicle environment takes place simultaneously.