Map Matching for Vehicle Localization Based on Serial Lidar Sensors

Abstract

For highly and fully automated driving an accurate vehicle localization is crucial. At any time the position has to be known within an accuracy of a few decimeters, which can not be guaranteed by single GNSS measurements. Urban canyons scenarios lead to situations where the satellite visibility is low and effects like multi-path occur. We suggest using LiDAR sensors to augment localization.We use a feature-based map matching approach to estimate the vehicle position and orientation. The approach firstly detects pole-like and vertical planar objects, curb stones and lane markings in the measurements of automotive LiDAR sensors, which are integrated as standard in modern vehicels. The detected features are matched to a reference HD map. They are combined in a filter approach, in addition to GPS/IMU and vehicle odometry measurements. We tested our approach in an urban scenario and on a highway. The resulting accuracy ranges from 0.24 m (2D) in challenging urban areas to 0.12 m in lateral direction on a highway. The lateral error on a highway is always sufficiently small to allow the driving lane to be correctly determined.