Probabilistic Integrating of 3D Building Models and GNSS for Reliable Vehicle Localization in Urban Areas—the GAIN Approach

S. Bauer, M. Obst, G. Wanielik

Abstract:	The awareness of the own vehicle position is a crucial requirement for numerous applications of intelligent transportation systems (ITSs) and advanced driver assistance systems (ADASs). In general, the aim on any localization approach is to determine a precise and reliable vehicle position based on sensor information like satellite observations which are subject to significant uncertainties. Satellite navigation is an interesting candidate to solve this task as it provides a straightforward solution for the absolute positioning task with decent performance under good conditions. Nevertheless, there are situations like in urban canyons where the classical satellite navigation approaches either fail or deliver inconsistent results. Especially the handling of non-line-of-sight (NLOS) measurements—which is also often described by the generic term multipath—is a challenging task. If such spoiled observations are not processed properly inside of the navigation algorithm, an unexpected bias might be introduced in the final user position which does not comply with the estimated confidence of the position. Recent research activities in the field of multi sensor and integrated navigation in GNSS-challenged environments show that by using multi-constellation geometries and incorporating additional information this problem can be mitigated up to certain extend. For example, the usage of 3D city models to describe the environment of the vehicle a promising and straightforward approach to perform consistency checks whether the direct line of sight between the receiver antenna and the satellite is present. However, most of the current approaches do assume high accurate 3D environmental digital maps to derive a discrete decisions about the LOS or NLOS case. Unfortunately, such map data is not available in general and the generation is expensive. An integrated approach which correctly handles the 3D maps by a probabilistic manner and directly integrates them into the localization equation has not been proposed up to now. The aim of this work is to further advance the authors' previously proposed generic online GNSS positioning algorithm for vehicular applications with integrated probabilistic multipath detection and mitigation. The probabilistic multipath algorithm is extended by a 3D environment model which is used to estimate the probability of line-of-sight (LOS) or non-line-of-sight (NLOS) reception. To do so, ray tracing is performed to calculate how close a direct signal reception path passes nearby buildings. As this has a high computational demand at the receiver side---which has to keep an up-to-date environment model available as well---a more efficient shadow map approach is evaluated as well. This shadow maps identify areas with NLOS reception conditions and can be computed offline at a service center to be shared (e.g. vehicle-to-infrastructure communication or 3G networks) and used by multiple receivers. Common algorithms usually alter measurement weightings somehow or exclude measurements using chosen thresholds. In contrast, both proposed models perform an integrated handling of probabilities (explicitly including the noise parameters of the digital maps) within the used Bayes Filter framework to autonomously increase localization accuracy and integrity as well. The proposed approach is directly working at the level of pseudoranges and can be plugged in as a first state into the typical refinement chain (SBAS, dead reckoning, map matching) used for positioning. The positioning algorithm described will be tested and validated with real data recorded during extensive test drives under typical urban conditions. To assess the absolute positioning performance, the results will be compared to state-of-the-art algorithms as well as a high reliable reference sensor system (NovAtel SPAN System with RTK and IMU). The evaluation includes a detailed analysis of the achieved accuracy as well as the provided integrity level. Furthermore, the results will be compared to classical RAIM solutions and post-processing methods like optimizers. Moreover, the multipath detection performance will be assessed in a software simulation, where multipath signals are generated for an urban scenario. Within this work, the multipath simulation software and its properties will be explained as well. Within this paper, an extension of previous work of the authors by introducing a probabilistic 3D environment model was proposed and validated within a simulation as well as by real-world data from extensive urban test drives. The algorithm was used in combination with a vehicular motion model to show respect to the physical constrains of typical legal manoeuvres. Furthermore, the concept of a shadow map as compact and efficient representation of the 3D model information appropriate for mobile communication was presented. It was shown, that by including additional uncertain information from a 3D environment model, the positioning accuracy can be increased by 10% compared to previous approaches. Furthermore, the improvement of the estimated integrity was proven and a proposal for appropriate potential ADAS applications was made.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	1267 - 1276
Cite this article:	Bauer, S., Obst, M., Wanielik, G., "Probabilistic Integrating of 3D Building Models and GNSS for Reliable Vehicle Localization in Urban Areas—the GAIN Approach," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 1267-1276.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In