Alpha-shape算法构建枣树点云三维模型

为了实现枣树智能化修剪作业，该研究提出了基于点云配准的自然光照环境下的果树三维重构方法，并针对传统最近点迭代（Iterative Closest Point，ICP）算法对待配准点云的空间位置要求苛刻的问题，提出了改进的点云配准算法。首先，使用彩色深度（RGB-D）相机采集不同角度下的枣树彩色和深度图像，并通过信息融合实现相应角度下的点云获取。其次，对点云进行背景去除和滤波处理，基于直方图设定分割阈值，提取单株枣树点云，并将放置在树根附近的标靶球作为标记，使用人工标记法进行两站点云初配准。最后，在初配准基础上计算点云的曲面法向量和曲率，由曲率相近的点构成配对点对，使用k维树最近点迭代（k dimensional-tree-Iterative Closest Point，kd-tree-ICP）算法完成精配准，对点云使用Alpha-shape算法面片化，实现表面重构。利用上述方法对多棵枣树进行全局配准并完整重构果树模型。试验结果表明，通过引入初配准，有效提高了点云配准的准确性和稳定性，配准误差均控制在1.0 cm以内，平均配准误差为0.78 cm；重构模型真实感较强，在外观上更加接近真实树，枝干相对误差控制在7%以内。该研究重构模型精度较高，可为枣树智能修剪提供可视化研究基础和技术支持。

Three-dimensional model construction method and experiment of jujube tree point cloud using Alpha-shape algorithm

Abstract: Jujube is widely cultivated in China because of their extremely high nutritional value and medical value. With the increase of the planting area, the shortcomings of manual management have become increasingly prominent, and it is urgent to realize the information management of jujube orchards. With the rise of smart agriculture, computer technology and agricultural production are combined to build digital agriculture. To realize the intelligent pruning of jujube trees, a three-dimensional reconstruction of jujube trees in Xinjiang was carried out. This study proposed a three-dimensional reconstruction method of fruit trees under a natural lighting environment based on point cloud registration. Moreover, aiming at the strict requirements of the spatial position of the registered point cloud, an improved point cloud registration algorithm was proposed based on the traditional Iterative Closest Point (ICP) algorithm for the strict requirements of the spatial position of the registered point cloud. Firstly, color images and depth images of fruit trees from different perspectives were collected by using an RGB-D camera, and point cloud acquisition under corresponding perspectives was achieved through information fusion. Secondly, Data preprocessing of fruit trees'' each piece of point cloud was carried out for background removing and original point cloud de-noising based on depth distance judgment and spare noise filtering methods respectively. The region of interest was extracted by setting the segmentation threshold based on the histogram of the number of point clouds, and accordingly, each relative accurate data set was obtained as the jujube tree''s point cloud in each specific perspective. Then, there were three target balls which were artificial markers near the root, and the artificial marking method was used to realize the initial cloud registration of the two sites. Finally, on the basis of initial registration, the surface normal vector and curvature of the point cloud were calculated, and the points with similar curvature formed a pair of points. The kd-tree was used to establish a high-dimensional index tree data structure to structure to decrease the cost of running time of point cloud registration. Then, the ICP algorithm was used to complete precise registration. The registered point cloud was triangulated using the Alpha-shape algorithm to achieve surface reconstruction. The above-mentioned methods of initial registration and precise registration were used to globally register multiple jujube trees and completely reconstruct the three-dimensional model of fruit trees. The experimental results showed that by introducing the initial registration, the accuracy and stability of the point cloud registration were effectively improved. The registration error was controlled within 1.0 cm, and the average registration error was 0.78 cm. The reconstructed model had a strong sense of reality and was closer to the real-world tree in appearance. The relative error between the ground truth of the branch and the reconstructed value was controlled within 7%, and the accuracy of the reconstructed model was higher. The reconstruction model had high accuracy, which could provide a visual research foundation and technical support for intelligent jujube pruning.