摘要
针对葡萄采摘机器人在自然环境下采摘点定位困难的问题,本研究提出了一种基于分水岭果梗图像分割和最小角度约束的采摘点定位方法。第一,将采集的葡萄果实图像变换到YUV颜色模型,对U通道灰度图进行均衡化处理,然后进行双阈值分割和形态学开操作去掉干扰区域,再对二值图像进行填充;第二,以检测到的葡萄果实位置为参考,确定葡萄串果梗感兴趣区域,对该感兴趣区域进行分水岭操作;第三,对果梗二值图像进行角点检测,然后分别对每个果梗检测到的角点数据进行线性回归,将拟合到的直线分别与垂直于地面直线进行角度计算,将夹角角度最小的拟合直线所在的果梗确定为葡萄串所连的果梗;第四,对该果梗检测到的角点数据进行K均值聚类分析,聚类中心为最佳采摘点。对采集于晴天顺光、晴天逆光和晴天遮阳这3种条件下各40幅夏黑葡萄图像进行验证,采摘点定位成功率为89.2%,单张葡萄图像的采摘点平均定位时间为0.65 s。说明该方法可以为葡萄采摘机器人提供准确的采摘点坐标信息。
In order to solve the difficulty of grape picking robot location in the natural environment,a method of picking point positioning was proposed based on watershed image segmentation and minimum angle constraint.Firstly,the collected grape images were transformed into the YUV color model,and histogram equalization was carried out for U channel grayscale image.Furthermore,double threshold segmentation and morphological operations were performed to remove the interference areas,and the binary image was filled.Secondly,using the detected location of grape fruit as reference,the region of interest(ROI)of grape stem was determined,and watershed operation was carried out.Thirdly,corner detection was performed on the binary image of grape fruit stem.Moreover,linear regression was performed on the corner points of each fruit stem,and the angles between the fitting straight lines and the line perpendicular to the ground were calculated,respectively.If the angle was the smallest,the fruit stalk was determined as the fruit stalk connected to the grape string.Finally,K-means clustering algorithm was performed on the corner points of the fruit stem,and the center of mass after clustering was used as the best picking point.In addition,40 images were collected for verification under three conditions:sunny day with direct sunlight,sunny day with backlighting and sunny day with shade.The results showed that the accuracy reached 89.2%under these three conditions,and the positioning time of single grape image was 0.65 s.In conclusion,the grape picking robot can obtain the exact coordinate of picking point by this method.
作者
雷旺雄
卢军
LEI Wang-xiong;LU Jun(College of Mechanical and Electrical Engineering, Shaanxi University of Science and Technology, Xi’an 710021, China)
出处
《江苏农业学报》
CSCD
北大核心
2020年第4期1015-1021,共7页
Jiangsu Journal of Agricultural Sciences
基金
陕西省科技厅工业攻关基金项目(2016GY-049)。
关键词
葡萄
采摘机器人
视觉定位
图像分割
分水岭
角点检测
K均值聚类
grape
picking robot
visual positioning
image segmentation
watershed
corner detection
K-means clustering
