An Approch for Automated Generation of High Precision Highway Maps

Abstract

Modern driver assistance systems increasingly support the driver to drive safer and more efficient. To fulfill these tasks information about the vehicle environment is essential. Besides extracting this information from sensors to perceive the environment, it is also possible to use stored static data. These models allow a reliable and fast access to the environmental data without the typical problems faced with the recognition by sensors. Nowadays street maps produced for navigation purposes reach a precision of several meters. To support the driver during manoeuvres however a precision one magnitude better is necessary. The positioning technology has steadily improved over the years, so it is possible today to determine a position with an accuracy of a few millimeters (i.e. using DGPS). With the introduction of Galileo, the quality and the level of service of future positioning solutions will be further improved. The high-precision positioning systems available today are for automotive applications too expensive. Due to the upcoming Galileo system, it can be expected that high-precision positioning will become ready for the mass market in the near future. As a result, high-precision street maps will also become more relevant in the future for advanced driver assistance systems. This paper describes a method that is able to generate largely automated street maps for highways which are usable as input for driver assistance systems supporting maneuvering tasks. The basis for the map generation is measured street data which is collected by the test vehicle ViewCar. This vehicle has a high-precision coupled navigation, which is based upon DGPS, an inertial measurement unit and an odometer. Furthermore the ViewCar has an optical lane detection system. By combining this information, the absolute position of the lane sign posting in world coordinates is calculated. Disturbances, errors and so on, resulting in a fluctuating data quality are reduced by taking several test sets from the same roadway lane. Along the sign posting, the median from the measurement values at constant distances is calculated to minimize these errors. The sign posting course calculated in this way shows a very high quality. Every test run results in a large amount of data hence the data handling is also a challenge. To overcome these challenges, all measurement data is stored in a spatially enabled database for efficient access in which the access is performed by location and not by time. The described procedure can be used to align all lanes on the road. To reduce this effort in one direction only one lane is measured and the other lanes are assumed to run parallel to the measured lane. Thus, only the lane widths of the other lanes must be determined. The lane widths of German freeways follow the guidelines from “Richtlinie für Anlage von Straßen – Querschnitt (RAS-Q)” sufficient accurately. Hence the widths of the lanes which are not directly measured are verified in one test drive. Afterwards the highway course is divided into sections and described with mathematical functions for a compact and exact representation. Several test drives have been performed on the freeway A2 to get data for the development of the method. The individual steps of the method have been implemented in software and tested on the collected data. The preliminary results show the general potential of the method.