Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Pandar128 dataset for lane line detection

Filip Beránek, Václav Diviš, Ivan Gruber

TL;DR

Pandar128 introduces the largest public LiDAR-based lane detection dataset, featuring 52,200 camera frames and 34,829 LiDAR frames captured with a 128-beam sensor, along with full calibration and synchronized odometry to support projection, fusion, and temporal modeling. It provides a lightweight baseline, SimpleLidarLane, that fuses BEV semantic segmentation, clustering, and RANSAC-based polyline fitting to produce interpretable lane representations. The paper also proposes IAM-F1, an interpolation-aware metric for polyline alignment in BEV space, and demonstrates that IAM-F1 outperforms standard metrics on a diverse test set. Together, these contributions enable reproducible, geometry-aware evaluation and practical LiDAR-based lane detection under varied driving conditions, with all data and code publicly released.

Abstract

We present Pandar128, the largest public dataset for lane line detection using a 128-beam LiDAR. It contains over 52,000 camera frames and 34,000 LiDAR scans, captured in diverse real-world conditions in Germany. The dataset includes full sensor calibration (intrinsics, extrinsics) and synchronized odometry, supporting tasks such as projection, fusion, and temporal modeling. To complement the dataset, we also introduce SimpleLidarLane, a light-weight baseline method for lane line reconstruction that combines BEV segmentation, clustering, and polyline fitting. Despite its simplicity, our method achieves strong performance under challenging various conditions (e.g., rain, sparse returns), showing that modular pipelines paired with high-quality data and principled evaluation can compete with more complex approaches. Furthermore, to address the lack of standardized evaluation, we propose a novel polyline-based metric - Interpolation-Aware Matching F1 (IAM-F1) - that employs interpolation-aware lateral matching in BEV space. All data and code are publicly released to support reproducibility in LiDAR-based lane detection.

Pandar128 dataset for lane line detection

TL;DR

Pandar128 introduces the largest public LiDAR-based lane detection dataset, featuring 52,200 camera frames and 34,829 LiDAR frames captured with a 128-beam sensor, along with full calibration and synchronized odometry to support projection, fusion, and temporal modeling. It provides a lightweight baseline, SimpleLidarLane, that fuses BEV semantic segmentation, clustering, and RANSAC-based polyline fitting to produce interpretable lane representations. The paper also proposes IAM-F1, an interpolation-aware metric for polyline alignment in BEV space, and demonstrates that IAM-F1 outperforms standard metrics on a diverse test set. Together, these contributions enable reproducible, geometry-aware evaluation and practical LiDAR-based lane detection under varied driving conditions, with all data and code publicly released.

Abstract

We present Pandar128, the largest public dataset for lane line detection using a 128-beam LiDAR. It contains over 52,000 camera frames and 34,000 LiDAR scans, captured in diverse real-world conditions in Germany. The dataset includes full sensor calibration (intrinsics, extrinsics) and synchronized odometry, supporting tasks such as projection, fusion, and temporal modeling. To complement the dataset, we also introduce SimpleLidarLane, a light-weight baseline method for lane line reconstruction that combines BEV segmentation, clustering, and polyline fitting. Despite its simplicity, our method achieves strong performance under challenging various conditions (e.g., rain, sparse returns), showing that modular pipelines paired with high-quality data and principled evaluation can compete with more complex approaches. Furthermore, to address the lack of standardized evaluation, we propose a novel polyline-based metric - Interpolation-Aware Matching F1 (IAM-F1) - that employs interpolation-aware lateral matching in BEV space. All data and code are publicly released to support reproducibility in LiDAR-based lane detection.

Paper Structure

This paper contains 19 sections, 2 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Datasets
  4. Lane Detection
  5. Data
  6. Sensor Setup
  7. Dataset collection
  8. Annotations
  9. Data Split Strategy
  10. Methods
  11. SimpleLidarLane: Baseline Extraction Method
  12. Evaluation Metrics
  13. Interpolation-Aware Matching F1 (IAM-F1)
  14. Standard Metrics
  15. Experiments
  16. ...and 4 more sections

Figures (7)

  • Figure 1: Image from the front-facing camera (left) and the corresponding LiDAR scan (right).
  • Figure 2: A) Example of point cloud semantic segmentation annotation and B) Example of polyline annotation.
  • Figure 3: Distribution of curvature (measured via Pearson coefficient) of annotated lane lines across frames.
  • Figure 4: Distribution of the number of annotated lane lines per frame across the dataset.
  • Figure 5: Architecture of the proposed SimpleLidarLane pipeline.
  • ...and 2 more figures