Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging GNSS and Onboard Visual Data from Consumer Vehicles for Robust Road Network Estimation

Balázs Opra, Betty Le Dem, Jeffrey M. Walls, Dimitar Lukarski, Cyrill Stachniss

TL;DR

This work tackles robust road graph extraction for autonomous navigation by fusing GNSS traces with basic vision data (lane markings and road boundaries) from consumer vehicle sensors. It reframes road graph construction as road centerline semantic segmentation using a CNN on rasterized, jointly informative inputs, followed by map-matching–driven refinement to produce a geometrically accurate and topologically coherent road network. The method, Probe2Road, employs a D-LinkNet-50 backbone trained from scratch with a connectivity-preserving loss, plus a sequence of postprocessing steps (skeletonization, gap filling, and intersection disambiguation) that leverage time-series trajectories to disambiguate stacked roads. Evaluations on multiple datasets (San Jose, San Francisco, Tokyo) with GEO and iTOPO metrics show superior performance over GNSS-only and vision-less baselines, with strong generalization to unseen regions and complex road topologies. The approach demonstrates robust automatic road graph inference using only standard fleet sensors, enabling scalable HD map generation for navigation and autonomous systems, and is recognized by the 2023 Woven by Toyota Invention Award.

Abstract

Maps are essential for diverse applications, such as vehicle navigation and autonomous robotics. Both require spatial models for effective route planning and localization. This paper addresses the challenge of road graph construction for autonomous vehicles. Despite recent advances, creating a road graph remains labor-intensive and has yet to achieve full automation. The goal of this paper is to generate such graphs automatically and accurately. Modern cars are equipped with onboard sensors used for today's advanced driver assistance systems like lane keeping. We propose using global navigation satellite system (GNSS) traces and basic image data acquired from these standard sensors in consumer vehicles to estimate road-level maps with minimal effort. We exploit the spatial information in the data by framing the problem as a road centerline semantic segmentation task using a convolutional neural network. We also utilize the data's time series nature to refine the neural network's output by using map matching. We implemented and evaluated our method using a fleet of real consumer vehicles, only using the deployed onboard sensors. Our evaluation demonstrates that our approach not only matches existing methods on simpler road configurations but also significantly outperforms them on more complex road geometries and topologies. This work received the 2023 Woven by Toyota Invention Award.

Leveraging GNSS and Onboard Visual Data from Consumer Vehicles for Robust Road Network Estimation

TL;DR

This work tackles robust road graph extraction for autonomous navigation by fusing GNSS traces with basic vision data (lane markings and road boundaries) from consumer vehicle sensors. It reframes road graph construction as road centerline semantic segmentation using a CNN on rasterized, jointly informative inputs, followed by map-matching–driven refinement to produce a geometrically accurate and topologically coherent road network. The method, Probe2Road, employs a D-LinkNet-50 backbone trained from scratch with a connectivity-preserving loss, plus a sequence of postprocessing steps (skeletonization, gap filling, and intersection disambiguation) that leverage time-series trajectories to disambiguate stacked roads. Evaluations on multiple datasets (San Jose, San Francisco, Tokyo) with GEO and iTOPO metrics show superior performance over GNSS-only and vision-less baselines, with strong generalization to unseen regions and complex road topologies. The approach demonstrates robust automatic road graph inference using only standard fleet sensors, enabling scalable HD map generation for navigation and autonomous systems, and is recognized by the 2023 Woven by Toyota Invention Award.

Abstract

Maps are essential for diverse applications, such as vehicle navigation and autonomous robotics. Both require spatial models for effective route planning and localization. This paper addresses the challenge of road graph construction for autonomous vehicles. Despite recent advances, creating a road graph remains labor-intensive and has yet to achieve full automation. The goal of this paper is to generate such graphs automatically and accurately. Modern cars are equipped with onboard sensors used for today's advanced driver assistance systems like lane keeping. We propose using global navigation satellite system (GNSS) traces and basic image data acquired from these standard sensors in consumer vehicles to estimate road-level maps with minimal effort. We exploit the spatial information in the data by framing the problem as a road centerline semantic segmentation task using a convolutional neural network. We also utilize the data's time series nature to refine the neural network's output by using map matching. We implemented and evaluated our method using a fleet of real consumer vehicles, only using the deployed onboard sensors. Our evaluation demonstrates that our approach not only matches existing methods on simpler road configurations but also significantly outperforms them on more complex road geometries and topologies. This work received the 2023 Woven by Toyota Invention Award.
Paper Structure (9 sections, 2 equations, 4 figures, 5 tables)

This paper contains 9 sections, 2 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Our Approach to Road Graph Inference
  4. Road Centerline Segmentation
  5. Road Centerline Refinement
  6. Experimental Evaluation
  7. Experimental Setup
  8. Performance of Our Approach
  9. Conclusion

Figures (4)

  • Figure 1: Left: Visualization of aligned fleet sensor data, red: road boundaries, green: lane markings, blue: GNSS trace points. Right: Inferred road network of sufficiently covered roads while ignoring those driven only once.
  • Figure 2: The D-LinkNet zhou2018cvprws model architecture used for road centerline segmentation from rasterized sensor data. Input data are the rasterized sensor data, namely traces (A), lane markers (B), and road boundaries (C).
  • Figure 3: Dataset coverage used for the experiments. Top left: San Jose, top right: San Francisco, bottom: Tokyo. Aggregated GNSS traces are shown in yellow. Purple, green, and orange tiles represent training, validation, and test tiles respectively.
  • Figure 4: Results from all methods on selected areas of the San Francisco and Tokyo test sets. OpenStreetMap backdrop provided for context.