Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Accurate Automatic 3D Annotation of Traffic Lights and Signs for Autonomous Driving

Sándor Kunsági-Máté, Levente Pető, Lehel Seres, Tamás Matuszka

TL;DR

The paper tackles the lack of large-scale, accurate 3D annotations for static traffic-management objects (traffic lights and road signs) in autonomous driving. It introduces a LiDAR-free pipeline that combines 2D detections from RGB images with ego-motion data to triangulate 3D centers, estimate bounding box extents and orientations, and suppress false positives through image-space re-projection, achieving localization errors around 0.2–0.3 m up to 200 m. A representative dataset of ~50,000 auto-annotated frames (≈320k lights, 550k signs) is released under CC BY-NC-SA 4.0, enabling training of image-based perception models for long-range traffic management objects. The approach demonstrates strong quantitative performance on Waymo and in-house datasets and provides a practical, scalable resource for non-commercial research that advances 3D perception without reliance on LiDAR.

Abstract

3D detection of traffic management objects, such as traffic lights and road signs, is vital for self-driving cars, particularly for address-to-address navigation where vehicles encounter numerous intersections with these static objects. This paper introduces a novel method for automatically generating accurate and temporally consistent 3D bounding box annotations for traffic lights and signs, effective up to a range of 200 meters. These annotations are suitable for training real-time models used in self-driving cars, which need a large amount of training data. The proposed method relies only on RGB images with 2D bounding boxes of traffic management objects, which can be automatically obtained using an off-the-shelf image-space detector neural network, along with GNSS/INS data, eliminating the need for LiDAR point cloud data.

Accurate Automatic 3D Annotation of Traffic Lights and Signs for Autonomous Driving

TL;DR

The paper tackles the lack of large-scale, accurate 3D annotations for static traffic-management objects (traffic lights and road signs) in autonomous driving. It introduces a LiDAR-free pipeline that combines 2D detections from RGB images with ego-motion data to triangulate 3D centers, estimate bounding box extents and orientations, and suppress false positives through image-space re-projection, achieving localization errors around 0.2–0.3 m up to 200 m. A representative dataset of ~50,000 auto-annotated frames (≈320k lights, 550k signs) is released under CC BY-NC-SA 4.0, enabling training of image-based perception models for long-range traffic management objects. The approach demonstrates strong quantitative performance on Waymo and in-house datasets and provides a practical, scalable resource for non-commercial research that advances 3D perception without reliance on LiDAR.

Abstract

3D detection of traffic management objects, such as traffic lights and road signs, is vital for self-driving cars, particularly for address-to-address navigation where vehicles encounter numerous intersections with these static objects. This paper introduces a novel method for automatically generating accurate and temporally consistent 3D bounding box annotations for traffic lights and signs, effective up to a range of 200 meters. These annotations are suitable for training real-time models used in self-driving cars, which need a large amount of training data. The proposed method relies only on RGB images with 2D bounding boxes of traffic management objects, which can be automatically obtained using an off-the-shelf image-space detector neural network, along with GNSS/INS data, eliminating the need for LiDAR point cloud data.
Paper Structure (15 sections, 16 figures, 4 tables)

This paper contains 15 sections, 16 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Automatic Annotation of Traffic Lights and Signs in 3D
  4. 3D Localization
  5. Extent Calculation
  6. Orientation Estimation
  7. Reducing False Positive Detections
  8. Classification of Object Attributes
  9. 3D Traffic Light and Road Sign Dataset
  10. Evaluation
  11. Validation Challenges
  12. Validation of the method on the Waymo dataset
  13. Validation of Automatic Traffic Sign Annotation on in-house dataset
  14. Validation of Automatic Traffic Light Annotation on in-house dataset
  15. Conclusion

Figures (16)

  • Figure 1: The main steps of the automatic annotation method.
  • Figure 2: Calculation of 3D bounding box center.
  • Figure 3: Calculation of 3D bounding box extent.
  • Figure 4: Calculation of 3D bounding box orientation.
  • Figure 5: Samples from the dataset with 3D traffic sign and light annotations. The bounding boxes are automatically generated by our method. Traffic light states are color-coded.
  • ...and 11 more figures