Abstract: The present disclosure relates to method and system for global localization using a combination of visual odometry and optimization approaches in order to map a vehicle trajectory to an offline map with a-priori knowledge of at least two observed streets in order to reduce the optimization parameter space from three dimensions to one dimensions compared to standard iterative closest point methods.

Abstract: A method and a device for evaluating image data of a vehicle camera, involves: a) recording a first image by the vehicle camera, b) recording a subsequent second image by the vehicle camera, c) extracting multiple correspondent features in the first and second images, d) assuming a motion of the vehicle camera (motion hypothesis) between the recordings of the first image and the second image, e) determining a reprojection error of an extracted feature, which indicates the difference between the flow measured from the feature extracted in the first and second images and the flow calculated from the motion hypothesis, and f) establishing outliers, wherein an extracted feature is established as an outlier if the reprojection error of this feature reaches or exceeds a threshold, wherein the threshold is not constant within an iteration that includes at least steps b) to g). This offers an improved detection and removal of outliers to estimate the motion of the vehicle camera.

Abstract: The present disclosure relates to method and system for global localization using a combination of visual odometry and optimization approaches in order to map a vehicle trajectory to an offline map with a-priori knowledge of at least two observed streets in order to reduce the optimization parameter space from three dimensions to one dimensions compared to standard iterative closest point methods.

Abstract: A method and a device for evaluating image data of a vehicle camera, involves: a) recording a first image by the vehicle camera, b) recording a subsequent second image by the vehicle camera, c) extracting multiple correspondent features in the first and second images, d) assuming a motion of the vehicle camera (motion hypothesis) between the recordings of the first image and the second image, e) determining a reprojection error of an extracted feature, which indicates the difference between the flow measured from the feature extracted in the first and second images and the flow calculated from the motion hypothesis, and f) establishing outliers, wherein an extracted feature is established as an outlier if the reprojection error of this feature reaches or exceeds a threshold, wherein the threshold is not constant within an iteration that includes at least steps b) to g). This offers an improved detection and removal of outliers to estimate the motion of the vehicle camera.