A Tree Point Cloud Simplification Method Based on FPFH Information Entropy

Abstract

LiDAR technology has been widely used in forest survey and research, but the high-resolution point cloud data generated by LiDAR equipment also pose challenges in storage and computing. To address this problem, we propose a point cloud simplification method for trees, which considers both higher similarity to the original point cloud and the area of the tree point cloud. The method first determines the optimal search neighborhood using the standard deviation of FPFH information entropy. Based on FPFH information entropy and Poisson disc sampling theory, the point cloud is partitioned and sampled. By optimizing the separation thresholds of significant feature points and less significant feature points using a genetic algorithm with the Hausdorff distance and point cloud area as the objective function, the final simplified point cloud is obtained. Validation with two point cloud data sets shows that the proposed method achieves good retention of the area information of the original point cloud while ensuring point cloud quality. The research provides new approaches and techniques for processing large-scale forest LiDAR scan point clouds, reducing storage and computing requirements. This can improve the efficiency of forest surveys and monitoring.

A new point cloud simplification algorithm based on V-P container constraint and normal vector angle information entropy

“…In this study, the algorithms mentioned in sections 2.3-2.6 were implemented using the programming language VC++ combined with the point cloud library program, and the proposed method was compared with traditional normal vector angle, method in [5], method in [28], and method in [30]. Each method adopted corresponding suitable optimal parameters.…”

“…The simplification effects of each method on the horse and elephant point cloud are shown in figures 7 and 8. (c) method in [5]; (d) method in [28]; (e) method in [30]. (e) method in [30].…”

“…As shown in figure 7, the traditional normal vector angle method exhibited a transitional simplification of the horse; the point cloud of the horse's abdomen was missing (red circle), and there were many redundant points in other areas. The method in [5] generally has a good simplification effect; however, when it comes to simplifying the front part of the horse, the degree of simplification of the rear part of the horse's abdomen is lower, and the simplification of the horse's feet is better. The method in [28] also has a good simplification effect on horses; however, there are still many redundant points in the abdomen and neck areas of the horse, and a point cloud is missing at a certain distance in the red circle.…”

“…Primary point cloud simplification methods were mainly based on grid layering, researchers have realized that the compressed point cloud should retain the points with feature information, which are called key points, this results in a simplified method based on preserving point cloud feature information. Many simplification methods preserve feature point information, by calculating the information entropy of each point [3][4][5][6] or curvature value [7,8] to compress the point cloud. The normal vector angle information entropy and curvature value of each point are the feature information of the point.…”

Point cloud simplification algorithm based on voxel convex hull

“…However, the authors did not verify the rapidity of this method, nor did they quantitatively evaluate the reduction effect. Both Li et al [14] and Hu et al [15] used fast point feature histogram to calculate features and simplified point cloud according to feature values. Yang et al [16] firstly calculated the curvature, angular entropy, density to form a multidimensional vector, and obtained the feature point according to the curvature.…”