Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Applying Extended Object Tracking for Self-Localization of Roadside Radar Sensors

Longfei Han, Qiuyu Xu, Klaus Kefferpütz, Gordon Elger, Jürgen Beyerer

TL;DR

This study tackles roadside radar self-localization in ITS by turning localization into a point-cloud registration problem anchored in vehicle trajectories. It combines Gaussian Process Extended Object Tracking (GP-EOT) to model moving vehicles as star-convex objects and produce labeled trajectory point clouds, with Semantic Iterative Closest Point (SICP) to register these clouds to aerial laser scan road maps, yielding a refined pose in SE(3) after fusing a coarse SE(3) initialization and a fine SE(2) alignment. Key contributions include (1) a GP-EOT-based trajectory framework that preserves shape and behavior labels, (2) a semantic ICP registration that leverages driving semantics, and (3) a data-efficient, sub-meter localization solution validated in both simulated and real-world deployments, outperforming prior methods in translational accuracy and reducing data requirements. The approach enables accurate roadside sensor localization, facilitating downstream traffic analysis and vehicle tracking under challenging urban conditions, with clear paths for extension to full 3D localization and lane-level semantic mapping.

Abstract

Intelligent Transportation Systems (ITS) can benefit from roadside 4D mmWave radar sensors for large-scale traffic monitoring due to their weatherproof functionality, long sensing range and low manufacturing cost. However, the localization method using external measurement devices has limitations in urban environments. Furthermore, if the sensor mount exhibits changes due to environmental influences, they cannot be corrected when the measurement is performed only during the installation. In this paper, we propose self-localization of roadside radar data using Extended Object Tracking (EOT). The method analyses both the tracked trajectories of the vehicles observed by the sensor and the aerial laser scan of city streets, assigns labels of driving behaviors such as "straight ahead", "left turn", "right turn" to trajectory sections and road segments, and performs Semantic Iterative Closest Points (SICP) algorithm to register the point cloud. The method exploits the result from a down stream task -- object tracking -- for localization. We demonstrate high accuracy in the sub-meter range along with very low orientation error. The method also shows good data efficiency. The evaluation is done in both simulation and real-world tests.

Applying Extended Object Tracking for Self-Localization of Roadside Radar Sensors

TL;DR

This study tackles roadside radar self-localization in ITS by turning localization into a point-cloud registration problem anchored in vehicle trajectories. It combines Gaussian Process Extended Object Tracking (GP-EOT) to model moving vehicles as star-convex objects and produce labeled trajectory point clouds, with Semantic Iterative Closest Point (SICP) to register these clouds to aerial laser scan road maps, yielding a refined pose in SE(3) after fusing a coarse SE(3) initialization and a fine SE(2) alignment. Key contributions include (1) a GP-EOT-based trajectory framework that preserves shape and behavior labels, (2) a semantic ICP registration that leverages driving semantics, and (3) a data-efficient, sub-meter localization solution validated in both simulated and real-world deployments, outperforming prior methods in translational accuracy and reducing data requirements. The approach enables accurate roadside sensor localization, facilitating downstream traffic analysis and vehicle tracking under challenging urban conditions, with clear paths for extension to full 3D localization and lane-level semantic mapping.

Abstract

Intelligent Transportation Systems (ITS) can benefit from roadside 4D mmWave radar sensors for large-scale traffic monitoring due to their weatherproof functionality, long sensing range and low manufacturing cost. However, the localization method using external measurement devices has limitations in urban environments. Furthermore, if the sensor mount exhibits changes due to environmental influences, they cannot be corrected when the measurement is performed only during the installation. In this paper, we propose self-localization of roadside radar data using Extended Object Tracking (EOT). The method analyses both the tracked trajectories of the vehicles observed by the sensor and the aerial laser scan of city streets, assigns labels of driving behaviors such as "straight ahead", "left turn", "right turn" to trajectory sections and road segments, and performs Semantic Iterative Closest Points (SICP) algorithm to register the point cloud. The method exploits the result from a down stream task -- object tracking -- for localization. We demonstrate high accuracy in the sub-meter range along with very low orientation error. The method also shows good data efficiency. The evaluation is done in both simulation and real-world tests.
Paper Structure (14 sections, 20 equations, 4 figures, 2 tables)

This paper contains 14 sections, 20 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Formulation
  4. Self-localization with accumulated dynamic point clouds
  5. Extended Object Tracking using Gaussian Processes
  6. Semantic Iterative Closest Point
  7. Implementation
  8. Generation of Target Point Cloud
  9. Generation of Source Point Cloud
  10. Evaluation
  11. Tracking Quality
  12. Localization of Radar Sensors
  13. Comparison with other method
  14. Conclusion

Figures (4)

  • Figure 1: Conceptual overview of the proposed localization method with trajectories collected using EOT. (a) Radar raw data and a coarse transformation matrix $\mathbf{T}_{coarse}$. (b) Accumulation of trajectories from different vehicle using EOT tracker. Blue, red and black are the "right turn", "left turn", "straight ahead" trajectories. (c) Preprocessing of aerial laser scan data to get labelled lanelets. Colors also represent driving behaviors. HD maps are used to crop out the road from the aerial laser scan. SICP algorithm is used to register the source point cloud to the target point cloud. (d) The resulting $\mathbf{T}_{fine}$ is used to calculate the sensor pose $\mathbf{T}_{\texttt{UTM}}$. Light green points are sensor data. Dark green and blue points are the aerial laser points.
  • Figure 2: Test scenarios. (a) Sensor S1 and S2 are placed in CARLA Simulation map Town 10. (b) S3 and S4 are placed in Ingolstadt Germany. The location symbols indicate the position of the sensor. The triangles depict the viewing direction and an approximate range of view. The image in (b) is retrieved from AerialImage
  • Figure 3: Example of result from GP-EOT and labels for indicating the driving behavior of the vehicle. (a) shows the result with measurement model from GP, where the sensor position is not considered in the measurement model. (b) shows the result with measurement model from GPimproved, with sensor position considered in the measurement model. The color indicates the driving behavior: right turn, left turn and straight ahead. The tracking starts at the position of the arrow.
  • Figure 4: Localization result of the four sensors. Both target and source points for SICP are shown in the figure. For radar point clouds, the blue points indicate left turning behavior, the red points indicate right turning behavior and the black points indicate driving straight ahead. The bottom right part extend the two