Radar-based Pose Optimization for HD Map Generation from Noisy Multi-Drive Vehicle Fleet Data

TL;DR

Improved poses are used to generate a global radar occupancy map, aimed to facilitate precise on-vehicle localization and can be used as a basis for an existing lane boundary map generation pipeline, majorly improving map output compared to its original pure line detection based optimization approach.

Abstract

High-definition (HD) maps are important for autonomous driving, but their manual generation and maintenance is very expensive. This motivates the usage of an automated map generation pipeline. Fleet vehicles provide sufficient sensors for map generation, but their measurements are less precise, introducing noise into the mapping pipeline. This work focuses on mitigating the localization noise component through aligning radar measurements in terms of raw radar point clouds of vehicle poses of different drives and performing pose graph optimization to produce a globally optimized solution between all drives present in the dataset. Improved poses are first used to generate a global radar occupancy map, aimed to facilitate precise on-vehicle localization. Through qualitative analysis we show contrast-rich feature clarity, focusing on omnipresent guardrail posts as the main feature type observable in the map. Second, the improved poses can be used as a basis for an existing lane boundary map generation pipeline, majorly improving map output compared to its original pure line detection based optimization approach.

Figures (7)

• Figure 1: Radar occupancy map over satellite image: Left with unaligned, right with aligned vehicle poses. The equally spaced posts of the middle guardrail section become clearly distinguishable after alignment.
• Figure 2: Comparison of evaluation metrics: mapping pipeline output (red), ground truth road boundary (green), and lane dividers (white, solid and dashed). The offset error is visualized in blue, the remaining error after offset correction, therefore the non-offset error, in yellow.
• Figure 3: Flowchart showing all components of the map generation process. The dataset is described in \ref{['subsec:dataset']}, correlation calculation based on grid-based fitting in \ref{['subsec:correlation']}, the pose graph optimization in \ref{['subsec:posegraph']}, the generation of the occupancy map in \ref{['subsec:occu']}, and of the lane marking based map in \ref{['subsec:lane']}.
• Figure 4: Occupancy maps of aligned point clouds. Only our grid-based fitting approach shows improvements over the unaligned point cloud, the others remain almost indistinguishable.
• Figure 5: Qualitative comparison between expectation maximization based lane boundary aligned (left) and radar aligned (right) lane mapping results. In green the road boundary, in white solid and dashed lane dividers. Satellite imagery is not accurately georeferenced. The radar aligned map appears more smooth and shows less artifacts.