Robust Normal Estimation for 3D LiDAR Point Clouds in Urban Environments.

Area of Science:

• Geospatial Science
• Computer Vision
• Remote Sensing

Background:

• Normal estimation is fundamental for processing light detection and ranging (LiDAR) data in urban environments.
• Existing methods struggle with noise and anisotropic sampling common in urban LiDAR datasets.
• Accurate normals are essential for tasks like building reconstruction and ground-cover classification.

Purpose of the Study:

• To develop a robust normal estimation method for urban LiDAR point clouds.
• To improve the accuracy of object segmentation and feature preservation in urban scenes.
• To address challenges posed by noise and varying point densities.

Main Methods:

• Constructing an octree-based hierarchical representation of LiDAR data.
• Detecting consistent neighborhoods across multiple scales.
• Leveraging the assumption of locally planar surfaces for regular urban objects.

Main Results:

• The method successfully estimates robust normals for various objects in urban environments.
• The estimated normals improve the accuracy of object segmentation and identification.
• Sharp features and complete outlines of objects are preserved, even with noise and anisotropic sampling.

Conclusions:

• The proposed octree-based method offers a robust solution for normal estimation in challenging urban LiDAR data.
• This technique enhances downstream processing tasks, leading to more accurate urban scene analysis.
• The method demonstrates superior performance compared to state-of-the-art approaches in experimental validations.