Multi-level height maps-based registration method for sparse LiDAR point clouds in an urban scene.

Area of Science:

• Robotics and Autonomous Systems
• Geospatial Data Processing
• Computer Vision

Background:

• LiDAR sensors are crucial for urban scene reconstruction.
• Current registration methods struggle with sparse, low-overlap LiDAR data.
• Urban LiDAR point clouds often share common flat ground surfaces.

Purpose of the Study:

• To develop a novel coarse registration method for LiDAR point clouds.
• To address challenges in finding stable 3D correspondences in sparse urban data.
• To improve the efficiency and accuracy of LiDAR point cloud registration.

Main Methods:

• Proposed a multi-level height (MH) maps-based coarse registration technique.
• Simplified 3D registration to 2D registration using MH maps.
• Extracted robust correspondences using height intervals and statistical height information.

Main Results:

• The MH maps method demonstrated stability and superior performance compared to state-of-the-art methods in sparse scenarios.
• Experiments using Livox Mid-100 and Velodyne HDL-64E LiDAR data validated the method's effectiveness.
• Runtime analysis confirmed the method's non-iterative nature results in faster processing.

Conclusions:

• The MH maps method offers a stable, accurate, and fast solution for LiDAR point cloud registration in urban environments.
• The approach effectively handles sparse data and can be extended to unordered multi-view point clouds.
• This method simplifies complex 3D registration tasks by leveraging common urban ground features.