欢迎您访问《智慧农业(中英文)》官方网站! English

Smart Agriculture ›› 2022, Vol. 4 ›› Issue (3): 12-23.doi: 10.12133/j.smartag.SA202207002

• 专刊--智慧果园关键技术与装备 • 上一篇    下一篇

基于激光点云的三维虚拟果园构建方法

冯涵1(), 张浩1, 王梓1, 江世界1, 刘伟洪1, 周凌卉2, 王亚雄2, 康峰2, 刘星星1, 郑永军1,3()   

  1. 1.中国农业大学 工学院,北京 100083
    2.北京林业大学 工学院,北京 100083
    3.现代农业装备与设施教育部工程研究中心,北京 100083
  • 收稿日期:2022-07-05 出版日期:2022-09-30 发布日期:2022-11-04
  • 基金资助:
    烟台市校地融合发展项目(2021XDRHXMPT29);国家重点研发计划项目(2018YFD0700603)
  • 作者简介:冯 涵(1998-),硕士研究生,研究方向为农业智能装备技术。E-mail:s20213071238@cau.edu.cn
  • 通讯作者: 郑永军 E-mail:s20213071238@cau.edu.cn;zyj@cau.edu.cn

Three-Dimensional Virtual Orchard Construction Method Based on Laser Point Cloud

FENG Han1(), ZHANG Hao1, WANG Zi1, JIANG Shijie1, LIU Weihong1, ZHOU Linghui2, WANG Yaxiong2, KANG Feng2, LIU Xingxing1, ZHENG Yongjun1,3()   

  1. 1.College of Engineering, China Agricultural University, Beijing 100083, China
    2.College of Engineering, Beijing Forestry University, Beijing 100083, China
    3.Modern Agricultural Equipment and Facilities Engineering Research Center, Beijing 100083, China
  • Received:2022-07-05 Online:2022-09-30 Published:2022-11-04
  • corresponding author: ZHENG Yongjun E-mail:s20213071238@cau.edu.cn;zyj@cau.edu.cn

摘要:

针对果园管理数字化程度低、构建方法较为单一等问题,本研究提出了一种基于激光点云的三维虚拟果园构建方法。首先采用手持式三维点云采集设备(3D-BOX)结合即时定位与地图构建-激光测距与测绘(Simultaneous Localization and Mapping-Lidar Odometry and Mapping,SLAM-LOAM)算法获取果园点云数据集;然后通过统计滤波算法完成点云数据离群点与噪声点的去除,并结合布料模拟算法(Cloth Simulation Filtering,CSF)与DBSCAN(Density-Based Spatial Clustering of Applications with Noise)聚类算法,实现地面去除与果树聚类分割,进而使用VoxelGrid滤波器降采样;最后利用Unity3D引擎,构建虚拟果园漫游场景,将作业机械的实时GPS(Global Positioning System)数据从WGS-84坐标系转换为高斯投影平面坐标系,并通过LineRenderer显示实时轨迹,实现作业机械运动轨迹控制与作业轨迹的可视化展示。为验证虚拟果园构建方法的有效性,在海棠果园与芒果园开展果园构建方法测试。结果表明,所提出的点云数据处理方法对海棠果树与芒果树聚类分割的准确率分别达到了95.3%与98.2%;通过与实际芒果园的果树行距、株距对比,虚拟芒果园的平均行间误差约为3.5%,平均株间误差约为6.6%。并且将Unity3D构建出的虚拟果园与实际果园相比,该方法能够有效复现果园三维实际情况,得到了较好的可视化效果,为果园的数字化建模与管理提供了一种技术方案。

关键词: 虚拟果园, 点云处理, Unity3D, 轨迹可视化, 激光雷达, SLAM-LOAM

Abstract:

To solve the problems of low level of digitalization of orchard management and relatively single construction method, a three-dimensional virtual orchard construction method based on laser point cloud was proposed in this research. First, the hand-held 3D point cloud acquistion equipment (3D-BOX) combined with the lidar odometry and mapping (SLAM-LOAM) algorithm was used to complete the acquisition of the point cloud data set of orchard; then the outliers and noise points of the point cloud data were removed by using the statistical filtering algorithm, which was based on the K-neighbor distance statistical method. To achieve this, a distance threshold model for removing noise points was established. When a discrete point exceeded, it would be marked as an outlier, and the point was separated from the point cloud dataset to achieve the effect of discrete point filtering. The VoxelGrid filter was used for down sampling, the cloth simulation filtering (CSF) cloth simulation algorithm was used to calculate the distance between the cloth grid points and the corresponding laser point cloud, and the distinction between ground points and non-ground points was achieved by dividing the distance threshold, and when combined with the density-based spatial clustering of applications with noise (DBSCAN) clustering algorithm, ground removal and cluster segmentation of orchard were realized; finally, the Unity3D engine was used to build a virtual orchard roaming scene, and convert the real-time GPS data of the operating equipment from the WGS-84 coordinate system to the Gauss projection plane coordinate system through Gaussian projection forward calculation. The real-time trajectory of the equipment was displayed through the LineRenderer, which realized the visual display of the motion trajectory control and operation trajectory of the working machine. In order to verify the effectiveness of the virtual orchard construction method, the test of orchard construction method was carried out in the Begonia fruit and the mango orchard. The results showed that the proposed point cloud data processing method could achieve the accuracy of cluster segmentation of Begonia fruit trees and mango trees 95.3% and 98.2%, respectively. Compared with the row spacing and plant spacing of fruit trees in the actual mango orchard, the average inter-row error of the virtual mango orchard was about 3.5%, and the average inter-plant error was about 6.6%. And compared the virtual orchard constructed by Unity3D with the actual orchard, the proposed method can effectively reproduce the actual three-dimensional situation of the orchard, and obtain a better visualization effect, which provides a technical solution for the digital modeling and management of the orchard.

Key words: virtual orchards, point cloud processing, Unity3D, trajectory visualization, LiDAR, SLAM-LOAM

中图分类号: