司徒伟熙1　魏宝源2　纪艺杭1　韩耀荣1　李震1,3,4　吕石磊1,3,4　宋淑然1,3　

薛秀云1,3,4

（1.华南农业大学电子工程学院（人工智能学院），广东 广州 510642

2.中山大学电子与信息工程学院（微电子学院），广东 广州 510275

3.国家柑橘产业技术体系机械研究室，广东 广州 510642

4.农业农村部华南热带智慧农业技术重点实验室，广东 广州 510642）

摘要：果树冠层参数直接反映果树的生长情况和产能潜力。针对单传感器点云密度低及传统算法精度低的问题，提出激光雷达与Kinect融合的果树冠层参数计算方法。首先，通过果树冠层检测系统获取果树冠层点云数据；然后，将果树冠层点云数据预处理后，通过采样一致性初始匹配（SAC-IA）和双向KD树改进的迭代最近点（KD-ICP）算法配准多传感器点云，改善点云位姿；接着，通过坐标转换融合两侧配准点云，获得果树冠层点云；最后，通过切片台体法计算果树冠层体积和叶面积。实验结果表明：该方法相较于传统的几何体拟合法、三维凸包法，果树冠层体积计算精度分别提升了38.12%、12.96%，叶面积计算精度分别提升了11.56%、2.78%；相较于Kinect和激光雷达的单传感器点云，融合点云的果树冠层体积计算精度分别提升了7.41%、12.62%，叶面积计算精度分别提升了19.41%、7.08%。该方法可准确计算果树冠层参数，为药肥精准变量喷施、果树估产等提供科学依据。

关键词：激光雷达；Kinect；点云融合；果树冠层；切片台体法

中图分类号：TP391;S24       文献标志码：A 文章编号：1674-2605(2023)03-0002-09

DOI：10.3969/j.issn.1674-2605.2023.03.002

Calculation Method of Fruit Tree Canopy Parameters Based on

Fusion of LiDAR and Kinect

SITU Weixi1  WEI Baoyuan2  JI Yihang1  HAN Yaorong1  LI Zhen1,3,4      

LYU Shilei1,3,4  SONG Shuran1,3  XUE Xiuyun1,3,4

(1.College of Electronic Engineering / College of Artificial Intelligence, South China Agricultural University, Guangzhou 510642, China  2.School of Electronics and Information Technology / School of Microelectronics, Sun Yat-sen University, Guangzhou 510275, China  3. Division of Citrus Machinery, China Agriculture Research System, Guangzhou 510642, China  4. Key Laboratory of Smart Agricultural Technology in Tropical South China, Ministry of Agriculture and Rural Affairs, P.R.China, Guangzhou 510642, China)

Abstract: The canopy parameters of fruit trees directly reflect their growth and productivity potential. A fruit tree canopy parameter calculation method based on the fusion of LiDAR and Kinect is proposed to address the issues of low density of single sensor point clouds and low accuracy of traditional algorithms. Firstly, obtain the point cloud data of fruit tree canopy through the fruit tree canopy detection system; Then, after preprocessing the fruit tree canopy point cloud data, multi-sensor point clouds are registered by sample consensus initial alignment (SAC-IA) and bidirectional KD-tree iterative closest point (KD-ICP) algorithm to improve the point cloud pose; Next, by integrating the registration point clouds on both sides through coordinate transformation, the fruit tree canopy point cloud is obtained; Finally, the canopy volume and leaf area of fruit trees were calculated using the slicing table method. The experimental results show that compared to the traditional geometric fitting method and three-dimensional convex hull method, this method has improved the accuracy of fruit tree canopy volume calculation by 38.12% and 12.96%, and the accuracy of leaf area calculation by 11.56% and 2.78%, respectively; Compared to the single sensor point clouds of Kinect and LiDAR, the accuracy of fruit tree canopy volume calculation using fused point clouds has been improved by 7.41% and 12.62%, respectively, and the accuracy of leaf area calculation has been improved by 19.41% and 7.08%, respectively. This method can accurately calculate the canopy parameters of fruit trees, providing scientific basis for precise variable spraying of drugs and fertilizers, and estimating fruit tree yield.

Keywords: LiDAR; Kinect; point cloud fusion; fruit tree canopy; slicing table method