High-accurate vehicle localization using digital maps and coherency images

Accurate, reliable, and affordable vehicle localization is one important task in current automotive research activities. It enables technologies like cooperative systems or enhanced map based assistance systems. There are a wide variety of approaches to reach this higher accuracy. The algorithm presented in this paper utilizes image landmarks in combination with a low-cost Global Navigation Satellite System (GNSS) receiver and vehicle odometry to achieve this. While similar approaches often extract features from camera images and match those features with map information, the algorithm presented in this work directly transforms map feature data, creating a image of map features, like the camera would see it. The evaluation of this image prediction uses the coherency value, which is derived from the structure tensor. By predicting the whole image, the incorporation of the map information is moved from feature level to signal level. The likelihood models used for the evaluation of the coherency image are derived from real, manually labeled data. We present promising results of a test drive in an area with complex intersections. Those results are compared to ground truth data.