Vieira Oliveira, Pedro Filipe (TNO), Smit, Robin (TNO Integrated Vehicle Safety), Alassi, Alaa (TNO), Brouwer, Jochem (TNO), Goos, Jorrit (TNO)

Robust Multi-Sensor Feature Matching for HD Map-Based Localization Using Graph Optimization

Scheduled for presentation during the Regular Session "S43c-Multi-Sensor Fusion and Perception for Robust Autonomous Driving" (FR-LA-T43), Friday, November 21, 2025, 16:20−16:40, Stradbroke

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

This work presents a robust multi-sensor localization framework that integrates feature matching with high-definition (HD) maps using graph-based optimization techniques. The proposed system leverages lane and pole detections extracted from LiDAR-camera and radar-camera (RACam) systems, alongside odometry derived from IMU and wheel encoders. A novel lane factor based on multivariate implicit line fitting is introduced, enabling accurate feature association even in complex road geometries. Furthermore, the robustness of the object detection stack is enhanced with PointPainting+, a refined feature projection algorithm that significantly reduces false positives caused by shadow points. The feature matcher is incorporated into a localization fusion module, closing the loop with a global EKF for improved accuracy and resilience to drift. Experimental results on Dutch highways demonstrate that our method reduces the lateral positioning error by more than 90% when compared to a localization fusion using GNSS. The proposed method achieves lane-level accuracy with lateral errors between ≈[0.06, 0.1] m with an accuracy of more than 94% within a 0.2 m bound, while maintaining real-time performance.