摘要
为解决整株盆栽金桔果实识别及总体计数问题,提出了基于三维点云配准的金桔果实识别方法。首先,使用RGB-D相机采集植物多角度点云数据并进行背景去除和去噪处理。然后采用随机采样一致性(Random sample consensus,RANSAC)算法进行圆柱拟合获得旋转中心轴参数,将点云绕中心轴旋转固定角度完成初配准,之后采用点到面的迭代最近点(Iterative closest point,ICP)算法完成精配准得到完整点云。最后,对点云进行欧氏聚类分割,采用随机采样一致性算法对聚类后点云进行球形分割,获得每个果实的三维空间位置并计数。本研究对9株盆栽金桔(共149个果实)进行识别,总计识别查全率为85.91%,查准率为79.01%,F1值为82.32%,果实数量预测值和真实值的决定系数为0.97,平均绝对百分比误差为16.02%。实验结果表明,本文方法不依赖颜色信息,能够有效识别整株植物中未成熟的青色果实,可为果实识别与产量估计等研究提供参考。
Kumquat is a kind of indoor ornamental plant that is deeply loved by consumers.The number and spatial distribution of its fruits are important indicators that determine the quality and sales price of kumquat.The recognition methods based on RGB images or single-view point clouds were difficult to accurately complete the calculation of the total fruits amount of the whole plant,and could not comprehensively display the three-dimensional spatial distribution of fruits.Therefore,a fruit recognition method based on point cloud registration was proposed to solve the problems of fruit recognition and total counting of the whole plant.Firstly,plants were placed on a rotating platform,and a low-cost RGB-D camera was used to collect the point cloud of plants at six angles for 60°every interval.The background was removed according to the spatial distance.The outlier noise was removed by radius filtering algorithm.The white color noise was removed based on the color information.And the“flying pixels”and edge noises were removed according to normal vector features and Euclidean clustering algorithm.Based on the random sampling consensus algorithm,the cylindrical point cloud of the rotating platform was segmented and the central axis was calculated.The point cloud was rotated around the central axis by a corresponding angle for initial registration.Then the point-to-plane ICP algorithm was used for accurate registration.Finally,Euclidean clustering algorithm was used to divide the plant point cloud into multiple clusters.And the spherical segmentation of each cluster was performed based on the random sampling consensus algorithm.The segmented spherical point clouds were the identified fruits,and its three-dimensional spatial distribution could be displayed according to the center and radius of the sphere.Totally nine potted kumquat plants(149 fruits in total)were identified in the fruit growing stage.The results showed that the total recall was 85.91%,precision was 79.01%and F1 value was 82.32%.Compared with the ground t
作者
朱启兵
张梦
刘振方
黄敏
李学成
ZHU Qibing;ZHANG Meng;LIU Zhenfang;HUANG Min;LI Xuecheng(Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, Jiangnan University, Wuxi 214122, China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2022年第5期209-216,共8页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家自然科学基金项目(61772240)。
关键词
盆栽金桔
果实识别
点云配准
点云分割
RGB-D相机
potted kumquat
fruits recognition
point cloud registration
point cloud segmentation
RGB-D camera