Forest point cloud denoising method based on curvature and density

When LiDAR scans woodlands, the three-dimensional (3D) point cloud data of woodlands directly collected usually contain a large number of noise points, which is caused by equipment performance, environmental interference, or other factors. The poor data quality of the point cloud data affects the accuracy of the extracted individual tree feature parameters and forest 3D reconstruction. To improve the quality of forest 3D point cloud data, a curvature- and density-based denoising method has been proposed in this study. First, the surface curvature features were calculated for the noise-dense trunk area of the forest for primary denoising, and the initial denoising results were used to construct the K-dimensional-Tree (KD-Tree) space index to search the spherical neighbourhood. The denoising was completed according to the proposed density calculation method (SN-NN, Spherical Neighbourhood and Nearest Neighbour). To evaluate the comprehensive performance of the proposed denoising method, a new denoising evaluation index has been designed by integrating the denoising precision and effective data retention degree of the forest point cloud data. The proposed method was validated based on forest point cloud data collected using Terrestrial LiDAR. The results show that the denoising precision and effective data retention degree of the proposed method are both preferable, which can effectively improve the quality of forest point cloud data.