Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging V2X for Collaborative HD Maps Construction Using Scene Graph Generation

Gamal Elghazaly, Raphael Frank

TL;DR

HDMapLaneNet tackles the real-time, collaborative construction of HD map geometry by leveraging V2X communication and scene graph generation. The method processes front-camera imagery with DeepLabv3 for features, uses DETR to detect lane centerlines as Bézier curves, and employs a Relational Graph Convolutional Network to predict connectivity, transmitting a GeoJSON graph to the cloud for global map fusion. On nuScenes, the approach demonstrates improved association prediction over a state-of-the-art baseline, highlighting the potential of vehicle-to-cloud collaboration to update localized HD map layers efficiently. This framework paves the way for scalable, real-time HD map maintenance without relying solely on dedicated mapping vehicles, with practical implications for autonomous navigation and safety in dynamic road environments.

Abstract

High-Definition (HD) maps play a crucial role in autonomous vehicle navigation, complementing onboard perception sensors for improved accuracy and safety. Traditional HD map generation relies on dedicated mapping vehicles, which are costly and fail to capture real-time infrastructure changes. This paper presents HDMapLaneNet, a novel framework leveraging V2X communication and Scene Graph Generation to collaboratively construct a localized geometric layer of HD maps. The approach extracts lane centerlines from front-facing camera images, represents them as graphs, and transmits the data for global aggregation to the cloud via V2X. Preliminary results on the nuScenes dataset demonstrate superior association prediction performance compared to a state-of-the-art method.

Leveraging V2X for Collaborative HD Maps Construction Using Scene Graph Generation

TL;DR

HDMapLaneNet tackles the real-time, collaborative construction of HD map geometry by leveraging V2X communication and scene graph generation. The method processes front-camera imagery with DeepLabv3 for features, uses DETR to detect lane centerlines as Bézier curves, and employs a Relational Graph Convolutional Network to predict connectivity, transmitting a GeoJSON graph to the cloud for global map fusion. On nuScenes, the approach demonstrates improved association prediction over a state-of-the-art baseline, highlighting the potential of vehicle-to-cloud collaboration to update localized HD map layers efficiently. This framework paves the way for scalable, real-time HD map maintenance without relying solely on dedicated mapping vehicles, with practical implications for autonomous navigation and safety in dynamic road environments.

Abstract

High-Definition (HD) maps play a crucial role in autonomous vehicle navigation, complementing onboard perception sensors for improved accuracy and safety. Traditional HD map generation relies on dedicated mapping vehicles, which are costly and fail to capture real-time infrastructure changes. This paper presents HDMapLaneNet, a novel framework leveraging V2X communication and Scene Graph Generation to collaboratively construct a localized geometric layer of HD maps. The approach extracts lane centerlines from front-facing camera images, represents them as graphs, and transmits the data for global aggregation to the cloud via V2X. Preliminary results on the nuScenes dataset demonstrate superior association prediction performance compared to a state-of-the-art method.

Paper Structure

This paper contains 16 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Framework
  3. Lane Representation
  4. Architecture
  5. DeepLabv3
  6. Detection Transformer
  7. Relational Graph Convolutional Network
  8. V2X
  9. Model Training
  10. Precision-Recall
  11. Detection Ratio
  12. Connectivity Precision-Recall
  13. Implementation and Main Results
  14. Implementation
  15. Main Results
  16. ...and 1 more sections

Figures (3)

  • Figure 1: The high-level architecture of HDMapLaneNet. The pipeline begins by processing camera images using DeepLabV3 deeplabv3 as an image-view feature extractor. Simultaneously, lane centerline segments are detected using DETR detr, which represents them as a Bézier curve. An RGCN RCGN then constructs a scene graph by modeling the connectivity between segments. Finally, the serialized scene graph is transmitted via V2X communication for aggregation and distribution.
  • Figure 2: Illustration of the Precision-Recall Metric.
  • Figure 3: Qualitative results on nuScenes dataset nuscenes.