Abstract

The point cloud-based 3D reconstruction techniques have been widely used in the tree phenotype monitoring, and the data denoising is the essential step due to the various noise problems encountered in practical applications. An effective and accurate denoising method for the point cloud of trees by the hybrid filtering scheme is proposed in this paper. The statistical filtering algorithm is firstly well studied to give the best parameters to remove the outliers of the point cloud data; and then an improved voxel filtering method is proposed by replacing the representative points of a voxel with the neighboring points of the center of gravity that is computed from the voxel in the original point cloud. By comparing the time cost and the spatial features to maintain, the experimental results fully prove the proposed method can effectively reduce the multiscale noise without damaging the geometric structure of the source data, performing better than the traditional voxel filter and the statistical filter.

中文翻译：

---

基于混合滤波方案的树木点云有效去噪方法

摘要

基于点云的3D重建技术已广泛应用于树木表型监测中，由于实际应用中遇到的各种噪声问题，数据去噪是必不可少的步骤。本文提出了一种基于混合滤波方案的树木点云有效且准确的去噪方法。首先深入研究统计滤波算法，给出最佳参数，去除点云数据的异常值；然后提出一种改进的体素滤波方法，通过用原始点云中的体素计算出的重心的邻近点来替换体素的代表点。通过比较时间成本和维护的空间特征，实验结果充分证明该方法能够在不破坏源数据几何结构的情况下有效降低多尺度噪声，性能优于传统体素滤波器和统计滤波器。