Lane Centerline Extraction Based on Surveyed Boundaries: An Efficient Approach Using Maximal Disks.

Area of Science:

• Computer Vision
• Robotics
• Geographic Information Systems

Background:

• Accurate road centerline extraction is vital for autonomous driving systems, aiding in trajectory prediction and planning.
• Current methods often struggle with sparse data, high computational costs, and limited accuracy in complex road layouts.
• Representing road maps compactly using sparse boundary points is advantageous for efficient data processing.

Purpose of the Study:

• To develop and evaluate a novel, accurate, and computationally efficient method for lane centerline extraction from sparse road boundary data.
• To compare the proposed method against existing techniques like Voronoi tessellation and distance transform for centerline extraction.
• To assess the performance of different centerline extraction approaches on both custom and public datasets.

Main Methods:

• Proposing a novel lane centerline extraction method based on identifying and linking internal maximal circles within lane boundaries.
• Evaluating the proposed method and comparing it with Voronoi tessellation and distance transform-based approaches.
• Utilizing both a self-created dataset and a public dataset for comprehensive performance evaluation.

Main Results:

• The proposed internal maximal circle linking method achieved a maximum deviation below 0.15 m and an RMSE of less than 0.01 m on a custom dataset.
• Existing methods showed significantly lower performance, with Voronoi tessellation yielding 1.7 m max deviation and 0.35 m RMSE, and distance transform yielding 1 m max deviation and 0.25 m RMSE.
• The novel approach demonstrates superior accuracy in extracting lane centerlines from sparse road boundary points.

Conclusions:

• The novel internal maximal circle linking method offers a significant improvement in accuracy for lane centerline extraction from sparse data.
• This method provides a more reliable foundation for trajectory prediction and planning in autonomous driving applications.
• The findings highlight the potential of geometric approaches for robust road feature extraction in challenging scenarios.