End-to-End Online Vectorized Map Construction With Confidence Estimates

This study proposes an end-to-end online vector map generation method, termed COMap, which integrates confidence estimation to enhance the robustness and reliability of environmental perception systems in autonomous driving. The method is based on the DETR framework and introduces a dynamic instance...

Full description

Saved in:
Bibliographic Details
Main Authors: Bo Huang, Feng Du, Tao Peng, Xiao-Long Zheng, Ji-Quan Wu, Jun-Hui Zhang
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11028039/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study proposes an end-to-end online vector map generation method, termed COMap, which integrates confidence estimation to enhance the robustness and reliability of environmental perception systems in autonomous driving. The method is based on the DETR framework and introduces a dynamic instance query mechanism to strengthen the geometric representation capabilities of map elements. Additionally, it employs a full-covariance Gaussian distribution for confidence modeling, effectively quantifying prediction uncertainty and thereby improving the reliability of the map data. Experimental results demonstrate that COMap achieves a higher mean Average Precision (mAP) value on the nuScenes dataset compared to existing methods and significantly reduces the false detection rate in complex scenarios, thereby enhancing the accuracy of road structure representation. This method not only improves the stability of online map generation but also provides more reliable data support for downstream tasks such as path planning and decision-making in autonomous driving.
ISSN:2169-3536