NELD-EC: Neighborhood-Effective-Line-Density-Based Euclidean Clustering for Point Cloud Segmentation

Area of Science:

• Computer Vision
• Robotics
• Geospatial Analysis

Background:

• Clustering lidar point clouds is challenging due to irregular shapes and varying densities.
• Existing Euclidean clustering algorithms struggle with noise and adaptive thresholding.

Purpose of the Study:

• To propose a novel Neighborhood Effective Line Density (NELD)-based Euclidean Clustering (NELD-EC) algorithm.
• To enhance the accuracy and robustness of lidar point cloud clustering.

Main Methods:

• Calculates Neighborhood Effective Line Density (NELD) to represent local point cloud density.
• Employs an adaptive distance threshold derived from local densities for clustering.
• Filters interfering points to refine neighborhood density calculations.

Main Results:

• NELD-EC demonstrates superior performance on simulated, fixed, and sequential point clouds.
• Requires simpler parameter tuning and is less sensitive to initial thresholds.
• Significantly reduces over-segmentation and under-segmentation errors.

Conclusions:

• NELD-EC offers improved stability and accuracy for lidar point cloud clustering.
• The algorithm is particularly well-suited for dynamic, complex environments and sequential data processing.