Skip to Main Content
This paper presents two variants of a Bayesian algorithm for vehicle localization which use vehicle motion data, a low-cost GNSS receiver, a gray scale camera, and different digital map data. The key idea of the algorithm is not to extract features like points or lines from the camera image for the Bayes update, but to predict entire images. While the first variant performs this image prediction based on explicit landmark information of a digital map, the second variant predicts camera images directly based on aerial images. In doing so, no conversion step from aerial images to feature maps is necessary. Finally, the paper presents results for both approaches based on extensive test drive data with highly accurate reference data.