基于视觉三维重构的室内玉米植株形态测量方法  

Research on Indoor Maize Plant Morphology Measurement MethodBased on Visual 3D Reconstruction

作　　者：马志艳[1,2] 万海迪 陈学海 王玉斌 周明刚 MA Zhiyan;WAN Haidi;CHEN Xuehai;WANG Yubin;ZHOU Minggang(School of Mechanical Engineering,Hubei Univ.of Tech.,Wuhan 430068,China;Hubei Province Agricultural Machinery Equipment Intelligent Engineering Technology Research Center,Wuhan 430068,China)

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068 [2]湖北省农机装备智能化工程技术研究中心,湖北武汉430068

出　　处：《湖北工业大学学报》2025年第1期1-6,共6页Journal of Hubei University of Technology

基　　金：国家重点研发计划子课题(2018YFD0701002-03)。

摘　　要：为实现高效、经济且可靠的作物形态测量,以玉米植株为对象,开发一种基于视觉三维重构的室内玉米植株形态测量方法。借助RGB-D双目相机从多角度采集玉米植株RGB图像及其对应的深度图像,获取目标植株多角度点云数据。经过点云的粗精配准、精简和滤波处理,成功重构了作物三维模型。在此基础上,通过目标分割和圆柱拟合技术,精确提取茎秆信息,从而测量作物株高和茎粗。进一步地,点云聚类技术去除茎秆,准确获取叶片数,并通过平面拟合技术测量叶夹角,采用贪婪投影三角化计算叶片面积,从而实现作物形态的测量。目标玉米植株测量实验结果表明:株高绝对误差为12.4 mm,相对误差为2.14%;最大茎粗平均相对误差为8.81%,最小茎粗平均相对误差为10.17%;聚类得到的叶片数量与实际叶片数一致;10个叶片叶夹角平均相对误差为7.65%,叶片面积平均相对误差为6.04%。这些结果均符合作物形态测量的技术要求,表明本方法能为玉米植株提供一种全面、定量化的评估和分析工具,以检测和记录作物的生长发育和表型特征。A method for indoor maize plant morphology measurement based on visual 3D reconstruction was studied,aiming to achieve efficient,low-cost,and reliable measurements.Firstly,RGB-D stereo cameras were used to capture depth images corresponding to the RGB images of maize plants from multiple angles,obtaining point cloud data of the target plants.The point cloud data underwent coarse and fine registration,simplification,and filtering to reconstruct the 3D model of the crops.Then,the model was segmented to extract stem segments using cylinder fitting,enabling the measurement of plant height and stem thickness.Clustering and segmentation were used to obtain the number of leaves by removing stem point cloud data.The leaf angle was measured by fitting the leaves to a plane,and the leaf area was calculated using greedy projection triangulation.The experimental results of maize plant measurement showed that the absolute error of plant height was 12.4mm,with a relative error of 2.14%.The average relative error of the maximum stem thickness was 8.81%,and the minimum stem thickness had an average relative error of 10.17%.The number of clustered leaves was equal to the actual number of leaves,and the average relative error of leaf angle for ten leaves was 7.65%.The average relative error of leaf area for ten leaves was 6.04%.The results met the requirements of crop morphology measurement technology,providing a comprehensive and quantitative method to evaluate and analyze the growth,development,and phenotypic characteristics of maize plants.

关 键 词：双目视觉 三维重构 点云配准 圆柱拟合 植物形态测量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]