GenMapping: Unveiling the Potential of Inverse Perspective Mapping for Robust Online HD Map Construction
IEEE Open Journal of Intelligent Transportation Systems · 2026 · 7 · 1492-1506
DOI: 10.1109/ojits.2026.3705400 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of Intelligent Transportation Systems, 2026. Verfasst von Siyu Li und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 55 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Special Funds for Construction of Innovative Provinces in Hunan Province, Hunan Provincial Research and Development Project.
Abstract (Originalsprache)
Online High-Definition (HD) maps learn unified bird’s-eye view layouts through multi-view sensor systems on intelligent vehicles, providing environmental priors for downstream tasks. However, current online HD map models, which embed parameters of visual sensors into training, exhibit poor generalization capability in different sensor configurations of vehicles. Inspired by the potential of Inverse Perspective Mapping (IPM), where camera parameters are decoupled from the training process, we have designed a universal map generation framework, GenMapping. The framework is established with a triadic synergy architecture, including principal and dual auxiliary branches. When faced with a coarse road image translated via IPM, the principal branch reconstructs robust features under the global state space models. The two auxiliary branches include the dense perspective and sparse prior branches. The former learns the correlation between static and moving objects, compensating for the missing information in IPM road images. The latter introduces OpenStreetMap (OSM) to provide robust prior knowledge for mapping. The triple-enhanced merging module is crafted to integrate the unique spatial features from three branches. To further improve generalization capabilities, a Cross-View Map Learning (CVML) scheme is leveraged to realize joint learning within the common space. Additionally, a Bidirectional Data Augmentation (BiDA) module is introduced to mitigate reliance on datasets concurrently. Experimental results demonstrate that the proposed method outperforms state-of-the-art approaches in both accuracy and inference speed. Moreover, across different vehicle sensor configurations, the generalization performance improves by 16.6% for semantic mapping and 12.1% for vectorized mapping. The source code will be publicly available at https://github.com/lynn-yu/GenMapping
Förderung
- National Natural Science Foundation of China
- Special Funds for Construction of Innovative Provinces in Hunan Province
- Hunan Provincial Research and Development Project
- State Key Laboratory of Autonomous Intelligent Unmanned Systems
Zitieren
APA
Li, S., Yang, K., Shi, H., Wang, S., Yao, Y., & Li, Z. (2026). GenMapping: Unveiling the Potential of Inverse Perspective Mapping for Robust Online HD Map Construction. IEEE Open Journal of Intelligent Transportation Systems, 7, 1492-1506. https://doi.org/10.1109/ojits.2026.3705400
Vancouver
Li S, Yang K, Shi H, Wang S, Yao Y, Li Z. GenMapping: Unveiling the Potential of Inverse Perspective Mapping for Robust Online HD Map Construction. IEEE Open Journal of Intelligent Transportation Systems. 2026;7:1492-1506. doi:10.1109/ojits.2026.3705400
BibTeX
@article{Li2026,
author = {Li, Siyu and Yang, Kailun and Shi, Hao and Wang, Song and Yao, You and Li, Zhiyong},
title = {GenMapping: Unveiling the Potential of Inverse Perspective Mapping for Robust Online HD Map Construction},
journal = {IEEE Open Journal of Intelligent Transportation Systems},
year = {2026},
volume = {7},
pages = {1492--1506},
doi = {10.1109/ojits.2026.3705400}
}
EUVIMED