LGmap: Local-to-Global Mapping Network for Online Long-Range Vectorized HD Map Construction

TL;DR

LGmap is an innovative online mapping system that leverages symmetric view transformation and hierarchical temporal fusion to construct long-range high-definition maps for autonomous driving, demonstrating high stability and efficiency.

Contribution

The paper introduces LGmap, a novel online mapping pipeline with symmetric view transformation and hierarchical temporal fusion modules for improved long-range HD map construction.

Findings

Achieves 0.66 UniScore on OpenLaneV2 test set.

Overcomes limitations of sparse feature representation.

Accelerates convergence with ped-crossing resampling.

Abstract

This report introduces the first-place winning solution for the Autonomous Grand Challenge 2024 - Mapless Driving. In this report, we introduce a novel online mapping pipeline LGmap, which adept at long-range temporal model. Firstly, we propose symmetric view transformation(SVT), a hybrid view transformation module. Our approach overcomes the limitations of forward sparse feature representation and utilizing depth perception and SD prior information. Secondly, we propose hierarchical temporal fusion(HTF) module. It employs temporal information from local to global, which empowers the construction of long-range HD map with high stability. Lastly, we propose a novel ped-crossing resampling. The simplified ped crossing representation accelerates the instance attention based decoder convergence performance. Our method achieves 0.66 UniScore in the Mapless Driving OpenLaneV2 test set.