Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HisTrackMap: Global Vectorized High-Definition Map Construction via History Map Tracking

Jing Yang, Sen Yang, Xiao Tan, Hanli Wang

TL;DR

This work tackles temporal inconsistencies in HD map construction for autonomous driving by introducing HisTrackMap, which maintains instance-level history maps $\mathcal{M}^t$ and uses Map-Trajectory Prior Fusion to provide geometry-aware priors for track queries. It couples PV and BEV feature sampling with history-based priors through cross-attention, enabling temporally smooth and accurate vectorized HD maps. Additionally, it proposes G-mAP, a global geometric evaluation metric that assesses both rasterized polygonal elements and vectorized polylines across sequences, addressing limitations of single-frame mAP and indirect consistency measures. Experiments on nuScenes and Argoverse2 show SOTA performance in both single-frame and temporal metrics, with faster training and robust behavior under localization perturbations, highlighting the practical impact for real-world map data collection and autonomous driving systems.

Abstract

As an essential component of autonomous driving systems, high-definition (HD) maps provide rich and precise environmental information for auto-driving scenarios; however, existing methods, which primarily rely on query-based detection frameworks to directly model map elements or implicitly propagate queries over time, often struggle to maintain consistent temporal perception outcomes. These inconsistencies pose significant challenges to the stability and reliability of real-world autonomous driving and map data collection systems. To address this limitation, we propose a novel end-to-end tracking framework for global map construction by temporally tracking map elements' historical trajectories. Firstly, instance-level historical rasterization map representation is designed to explicitly store previous perception results, which can control and maintain different global instances' history information in a fine-grained way. Secondly, we introduce a Map-Trajectory Prior Fusion module within this tracking framework, leveraging historical priors for tracked instances to improve temporal smoothness and continuity. Thirdly, we propose a global perspective metric to evaluate the quality of temporal geometry construction in HD maps, filling the gap in current metrics for assessing global geometric perception results. Substantial experiments on the nuScenes and Argoverse2 datasets demonstrate that the proposed method outperforms state-of-the-art (SOTA) methods in both single-frame and temporal metrics. The project page is available at: https://yj772881654.github.io/HisTrackMap.

HisTrackMap: Global Vectorized High-Definition Map Construction via History Map Tracking

TL;DR

This work tackles temporal inconsistencies in HD map construction for autonomous driving by introducing HisTrackMap, which maintains instance-level history maps and uses Map-Trajectory Prior Fusion to provide geometry-aware priors for track queries. It couples PV and BEV feature sampling with history-based priors through cross-attention, enabling temporally smooth and accurate vectorized HD maps. Additionally, it proposes G-mAP, a global geometric evaluation metric that assesses both rasterized polygonal elements and vectorized polylines across sequences, addressing limitations of single-frame mAP and indirect consistency measures. Experiments on nuScenes and Argoverse2 show SOTA performance in both single-frame and temporal metrics, with faster training and robust behavior under localization perturbations, highlighting the practical impact for real-world map data collection and autonomous driving systems.

Abstract

As an essential component of autonomous driving systems, high-definition (HD) maps provide rich and precise environmental information for auto-driving scenarios; however, existing methods, which primarily rely on query-based detection frameworks to directly model map elements or implicitly propagate queries over time, often struggle to maintain consistent temporal perception outcomes. These inconsistencies pose significant challenges to the stability and reliability of real-world autonomous driving and map data collection systems. To address this limitation, we propose a novel end-to-end tracking framework for global map construction by temporally tracking map elements' historical trajectories. Firstly, instance-level historical rasterization map representation is designed to explicitly store previous perception results, which can control and maintain different global instances' history information in a fine-grained way. Secondly, we introduce a Map-Trajectory Prior Fusion module within this tracking framework, leveraging historical priors for tracked instances to improve temporal smoothness and continuity. Thirdly, we propose a global perspective metric to evaluate the quality of temporal geometry construction in HD maps, filling the gap in current metrics for assessing global geometric perception results. Substantial experiments on the nuScenes and Argoverse2 datasets demonstrate that the proposed method outperforms state-of-the-art (SOTA) methods in both single-frame and temporal metrics. The project page is available at: https://yj772881654.github.io/HisTrackMap.

Paper Structure

This paper contains 13 sections, 2 equations, 2 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. HisTrackMap
  4. Overview Architecture
  5. Instance-Level History Maps
  6. Map-Trajectory Prior Fusion
  7. Global Geometric HD Mapping Evaluation
  8. Experiments
  9. Experimental Settings
  10. Comparision with SOTA Method
  11. Quantitative Evaluations
  12. Ablation and Robustness Studies
  13. Conclusion

Figures (2)

  • Figure 1: The global rasterized map with history maps. Red and green represent divider and boundary categories, respectively. In the history maps, different colors indicate unique tracked map instances. The pink instance is tracked at $\mathbf{T}_1$, consistently propagating temporal geometric information and providing priors for the corresponding track query at $\mathbf{T}_2$.
  • Figure 2: Qualitative visualization on nuScenes val set.