B$^2$F-Map: Crowd-Sourced Mapping with Bayesian B-Spline Fusion
Yiping Xie, Yuxuan Xia, Erik Stenborg, Junsheng Fu, Axel Beauvisage, Gabriel E. Garcia, Tianyu Wu, Gustaf Hendeby
AI summary
Problem
Existing crowd-sourced mapping methods either rely on prior high-definition maps or ignore localization and mapping uncertainties during fusion, preventing scalable and cost-effective map updates.
Approach
The authors propose B2F-Map, a pipeline that performs on-vehicle localization and multi-lane tracking with a PMB filter, followed by on-cloud pose graph optimization and a density-invariant Bayesian fusion algorithm for B-spline trajectories.
Key results
- Bandwidth-efficient pipeline combining on-vehicle tracking and on-cloud optimization
- Novel Bayesian fusion method for B-spline trajectories with varying control point densities
- Demonstrated geometric consistency on 70 km of real-world crowd-sourced data
- Open-source dataset and code release for community validation
Why it matters
Enables scalable, low-cost HD map generation for autonomous driving using widely available production vehicles instead of expensive survey fleets.
Abstract
Crowd-sourced mapping offers a scalable alterna- tive to creating maps using traditional survey vehicles. Yet, ex- isting methods either rely on prior high-definition (HD) maps or neglect uncertainties in the map fusion. In this work, we present a complete pipeline for HD map generation using production ve- hicles equipped only with a monocular camera, consumer-grade GNSS, and IMU. Our approach includes on-cloud localization using lightweight standard-definition maps, on-vehicle mapping via an extended object trajectory (EOT) Poisson multi-Bernoulli (PMB) filter with Gibbs sampling, and on-cloud multi-drive optimization and Bayesian map fusion. We represent the lane lines using B-splines, where each B-spline is parameterized by a sequence of Gaussian distributed control points, and propose a novel Bayesian fusion framework for B-spline trajectories with differing density representation, enabling principled handling of uncertainties. We evaluate our proposed approach, B2F- Map, on large-scale real-world datasets collected across diverse driving conditions and demonstrate that our method is able to produce geometrically consistent lane-level maps.