Ufil: A Unified Framework for Infrastructure-based Localization

TL;DR

Ufil is an open-source, standardized framework that unifies infrastructure-based localization components, enabling flexible integration of heterogeneous data sources for accurate, real-time vehicle localization.

Contribution

The paper introduces Ufil, a modular, reusable, open-source framework with reference implementations that simplifies the development of infrastructure-based localization systems.

Findings

Ufil successfully integrates multiple data sources into a single localization pipeline.

The system achieves lane-level lateral accuracy with RMSE around 0.3 meters.

Median latency from sensing to output remains below 100 ms.

Abstract

Infrastructure-based localization enhances road safety and traffic management by providing state estimates of road users. Development is hindered by fragmented, application-specific stacks that tightly couple perception, tracking, and middleware. We introduce Ufil, a Unified Framework for Infrastructure-Based Localization with a standardized object model and reusable multi-object tracking components. Ufil offers interfaces and reference implementations for prediction, detection, association, state update, and track management, allowing researchers to improve components without reimplementing the pipeline. Ufil is open-source C++/ROS 2 software with documentation and executable examples. We demonstrate Ufil by integrating three heterogeneous data sources into a single localization pipeline combining (i) vehicle onboard units broadcasting ETSI ITS-G5 Cooperative Awareness Messages, (ii) a lidar-based roadside sensor node, and (iii) an in-road sensitive surface layer. The pipeline runs unchanged in the CARLA simulator and a small-scale CAV testbed, demonstrating Ufil's scale-independent execution model. In a three-lane highway scenario with 423 and 355 vehicles in simulation and testbed, respectively, the fused system achieves lane-level lateral accuracy with mean lateral position RMSEs of 0.31 m in CARLA and 0.29 m in the CPM Lab, and mean absolute orientation errors around 2.2{\deg}. Median end-to-end latencies from sensing to fused output remain below 100 ms across all modalities in both environments.