Automated Extraction of Driving Lines from Mobile Laser Scanning Point Clouds

Lingfei Ma, Tianyu Wu, Ying Li, Jonathan Li, Yiping Chen, Michael A. ChapmanAdvances 2019 · DOI

Abstract

This paper presents a novel approach to automated generation of driving lines from mobile laser scanning (MLS) point cloud data. The proposed method consists of three steps: road surface segmentation, road marking extraction and classification, and driving line generation. The voxel-based upward-growing algorithm was firstly used to extract ground points from the raw MLS point clouds followed by segmentation of road surface using a region-growing algorithm. Then, the statistical outlier removal filter was applied to separate and refine the road marking points followed by extracting and classifying the lane markings based on the geometric features of different road markings using empirical hierarchical decision trees. Finally, land node structures were constructed followed by generation of driving lines using a curve-fitting algorithm. The proposed method was tested on both circular road sections and irregular intersections. The overall performance of the proposed algorithms is promising with 96.0% recall, 100.0% precision, and 98.0% F1-score for the lane marking extraction specifically. Validation results demonstrate that the driving lines can be effectively generated within 20 cm-level localization accuracy, meeting the requirements for fully autonomous driving functions.

Introduction

For an autonomous vehicle (AV) to navigate safely, it requires high-definition (HD) maps with centimeter-level precision. Specifically, AVs need "driving lines"—the precise paths within a lane that a car should follow. While Mobile Laser Scanning (MLS) systems can collect massive amounts of 3D data quickly, turning that "cloud" of millions of points into a clean, machine-readable map is a major bottleneck. The data is often unordered, noisy, and varies in intensity. Extracting road markings and generating smooth paths through complex areas like irregular intersections or sharp curves remains a significant technical challenge. This paper introduces a semi-automated pipeline to transform raw, noisy MLS point clouds into these critical HD driving lines.

Methodology

The authors propose a three-stage workflow to automate the extraction process, as illustrated in the flowchart below.

Flowchart of the proposed three-stage pipeline for driving line generation.

1. Road Surface Segmentation

To handle massive data volumes, the team utilized a voxel-based upward-growing algorithm. This partitions the 3D space into cubes (voxels) and identifies ground points based on height differences. By processing data in voxels rather than individual points, the computational burden is significantly reduced. A subsequent region-growing algorithm then isolates the actual road surface from curbs and sidewalks.

Voxel-based upward-growing algorithm details: (a) road partitioning, (b) voxelized point cloud, (c) neighborhood search, and (d) height difference checks.

2. Road Marking Extraction & Classification

Using the intensity of laser returns—since road paint is more reflective than asphalt—the system applies Otsu’s thresholding to isolate markings. These markings are then classified into categories (e.g., zebra crossings, boundary lines, arrows) using empirical hierarchical decision trees based on geometric features like length and width.

3. Driving Line Generation

The system constructs a "line node structure" using trajectory data. To ensure paths are smooth enough for vehicle navigation, smoothing spline curve fitting is used for circular roads, while Catmull-Rom splines are applied at complex intersections. The latter is particularly effective because it passes through all control points, ensuring the generated path stays strictly within lane boundaries.

Results and Discussion

The study utilized two distinct datasets collected in Xiamen, China, using a RIEGL VMX-450 MLS system: a 70.5-meter circular road section (21 million points) and a complex intersection (22 million points).

The proposed method showed high reliability across both environments:

  • Lane Marking Extraction: Achieved 96.0% recall and 100.0% precision (98.0% F1-score).
  • Classification Accuracy: Overall accuracy for classifying different types of markings ranged between 81.98% and 89.26%.
  • Localization Precision: Driving lines were generated with 20 cm-level accuracy. In the 20 cm buffer zone, the "miscoding" rate (the rate of the line falling outside the acceptable boundary) was only 3.5% on average.

Conclusion

The results demonstrate that HD maps must transition from being "human-readable" visual markings to "machine-readable" mathematical driving lines. While geometric decision trees worked well for standard markings, they struggled with complex indicators like text or HOV diamonds. Future work will focus on integrating deep learning-based methods to recognize the semantic meaning of complex urban features, further reducing human intervention in HD map production.

Summary updated

Abstract

This paper presents a novel approach to automated generation of driving lines from mobile laser scanning (MLS) point cloud data. The proposed method consists of three steps: road surface segmentation, road marking extraction and classification, and driving line generation. The voxel-based upward-growing algorithm was firstly used to extract ground points from the raw MLS point clouds followed by segmentation of road surface using a region-growing algorithm. Then, the statistical outlier removal filter was applied to separate and refine the road marking points followed by extracting and classifying the lane markings based on the geometric features of different road markings using empirical hierarchical decision trees. Finally, land node structures were constructed followed by generation of driving lines using a curve-fitting algorithm. The proposed method was tested on both circular road sections and irregular intersections. The smoothing spline curve fitting model was tested on the circular road sections, while the Catmull-Rom spline with five control points was used to generate the driving lines at road intersections. The overall performance of the proposed algorithms is promising with 96.0% recall, 100.0% precision, and 98.0% F1-score for the lane marking extraction specifically. Most significantly, the validation results demonstrate that the driving lines can be effectively generated within 20 cm-level localization accuracy at an average of 3.5% miscoding using MLS point clouds, which meets the requirement of localization accuracy of fully autonomous driving functions. The results demonstrate the proposed methods can successfully generate road driving lines in the test datasets to support the development of high-definition maps.
BibTeX
@inproceedings{ma2019automated,
  author = {Lingfei Ma and Tianyu Wu and Ying Li and Jonathan Li and Yiping Chen and Michael A. Chapman},
  title = {{Automated Extraction of Driving Lines from Mobile Laser Scanning Point Clouds}},
  booktitle = {Advances},
  year = {2019},
  doi = {10.5194/ica-adv-1-12-2019},
}
Figure