High definition map-based vehicle localization for highly automated driving: Geometric analysis

Abstract

High definition (HD) map is a critical part of highly automated driving (HAD) technology and shows potential for high precision vehicle localization when GNSS signals are not available. The current study of using HD map for localiz ation is mostly based on Simultaneous Localization and Mapping (SLAM) technique, which requires high computing power, huge storage space, and quick data transmission ability. Therefore, a study of a new HD map based vehicle localization method which requires less computation is necessary. Geometry is one key component that affects the quality of localization, including accuracy, reliability, and separability. Analysing the geometry can provide reference for designing a localization system to meet the quality requirement of HAD, but is rarely studied. This paper aims to design a high precision and reliable localization system using HD map as a sensor, and the influence of geometry is also explored. Geometric strength is evaluated under different scenarios considering three factors, including feature distribution type, feature number, and distance between vehicle and feature. The results show Minimum Detectable Bias (MDB) and Minimal Separable Bias (MSB) are mostly affected by feature number and distance between vehicle and feature. Randomly distribution, more detected features and close distance between the host vehicle and the features may all contribute to good quality of vehicle position estimation.