LiDAR point cloud down-sampling strategy with multi-dimensional membership fusion

This study devises an innovative LiDAR point cloud down-sampling strategy that capitalizes on the properties of Fuzzy C Means (FCM) clustering membership functions in each dimension. Traditional down-sampling methods frequently encounter difficulties in striking a balance between computational efficiency and feature preservation, particularly for large-scale datasets. To tackle this issue, our approach breaks down the three-dimensional simplification problem into independent one-dimensional analyses. Specifically, FCM clustering is carried out separately on the X, Y, and Z coordinates to generate dimension-wise membership functions. These functions are then intelligently integrated to calculate comprehensive importance scores for each point, facilitating adaptive sampling that eliminates redundant data while retaining critical geometric features. Experimental results demonstrate that our method outperforms conventional approaches, including voxel grid, random, and farthest point sampling, in terms of geometric fidelity. The proposed method shows strong potential for real-time applications involving large-scale point clouds in fields such as autonomous driving, robotic navigation, and 3D reconstruction.