摘要
为了提高农业视觉导航系统对作物定位的精确性,提出了一种基于双目视觉的作物点云获取与分割定位方法。该方法采用ZED双目相机采集作物左右视图,通过视差原理获取作物的3D点云数据,利用点云离散程度和体素化网格方法对初始点云数据的离散点和冗余数据进行去除,然后在预处理后的点云图中利用基于点云法线角度差的区域生长分割出每株作物的点云簇,用每个点云簇中所有点的平均坐标值作为该株作物的三维坐标,结合视觉系统坐标系,计算出作物与相机的水平距离以及水平偏角,从而实现作物定位。试验结果表明,该方法测得的作物平均距离误差为1.89%,平均角度误差为2.17%,该算法可以对作物进行准确定位,为基于双目视觉导航的路径规划提供可靠的定位信息。
In order to improve the accuracy of crop positioning in agricultural visual navigation systems, a point cloud acquisition and segmentation and location method of crops based on binocular vision was proposed in this study. The left and right view images were taken by a ZED binocular camera, and 3D point cloud data of crops was obtained based on the parallax principle. Then, the outliers and redundant data of the initial point cloud data were removed by the point cloud dispersion degree and voxelization grid method, respectively. After that, region growth segmentation based on point cloud normal angle difference was used to segment crop point cloud clusters, and the average coordinate value of all points in each point cloud cluster was taken as the three-dimensional coordinate of this plant. Combined with coordinate system of visual system, the horizontal distance between the crop and the camera and horizontal angle were calculated, which could provide location information on the distance and direction of the crop relative to the machinery. Experimental results showed that the average distance error of crops measured by this method was 1.89%, and the average angle error was 2.17%. This algorithm can locate crops accurately and provide reliable location information for the subsequent path planning based on binocular visual navigation.
作者
廖娟
汪鹞
尹俊楠
刘路
张顺
朱德泉
LIAO Juan;WANG Yao;YIN Jun-nan;LIU Lu;ZHANG Shun;ZHU De-quan(School of Engineering, Anhui Agricultural University, Hefei 230036, China)
出处
《江苏农业学报》
CSCD
北大核心
2019年第4期847-852,共6页
Jiangsu Journal of Agricultural Sciences
基金
国家重点研发计划项目(2018YFD0700304)
安徽省自然科学基金项目(1708085QF148)
安徽农业大学青年基金项目(2016ZR008)
关键词
双目视觉
ZED相机
作物定位
3D点云
点云分割
binocular vision
ZED camera
crop location
3D point cloud
point cloud segmentation
