| 基于激光雷达的农业机器人果园树干检测算法 |
| |
| 引用本文： | 牛润新,张向阳,王杰,祝辉,黄健,陈正伟.基于激光雷达的农业机器人果园树干检测算法[J].农业机械学报,2020,51(11):21-27. |
| |
| 作者姓名： | 牛润新 张向阳 王杰 祝辉 黄健 陈正伟 |
| |
| 作者单位： | 中国科学院合肥物质科学研究院,合肥230088;中国科学技术大学,合肥230026;中国科学院合肥物质科学研究院,合肥230088 |
| |
| 基金项目： | 国家重点研发计划项目（2018YFD07000602、2016YFD0701401、2017YFD07000303）、中国科学院青促会项目（2017488）、中国科学院135项目（KP-2017-35、KP-2017-13、KP-2019-16）、中国科学院机器人与智能制造创新研究所自主研究项目（C2018005）和安徽省新能源汽车暨智能网联汽车产业技术创新工程项目 |
| |
| 摘 要： | 针对丘陵山区果园中的斜坡及杂草影响果树检测精度的问题，提出了一种基于激光雷达的树干检测算法。首先，利用单线激光雷达获取环境信息，通过数据预处理滤除噪声点及无法利用的数据点，以树干为目标设定聚类半径，根据数据点到激光雷达的距离自适应设定聚类阈值，完成初步聚类；然后，利用初步聚类结果及地面类内数据点量大、且大致呈一条直线的特征，将数据点超过一定数量的类进行二次曲线拟合，将拟合半径大于一定阈值的类视为地面干扰，并将其剔除；最后，利用杂草枝叶类中数据点之间距离不连续的特征，将存在一定数量的相邻数据点距离较大的类视为杂草枝叶类，并将其剔除，从而完成对果园中果树树干的检测。结果表明：在无干扰情况下，对树干的误检率为0.76%、漏检率为1.90%，平均正确率为97.3%；在只存在地面干扰的情况下，树干检测平均正确率为96.1%；在只存在杂草干扰的情况下，树干检测平均正确率为91.4%；在同时存在地面和杂草干扰的情况下，树干检测平均正确率为91.9%，综合以上各种情况的树干检测平均正确率为95.5%，该方法可用于丘陵山区树干较明显的乔化果园中的树干检测，为精准农业装备在丘陵山区果园中的导航应用提供参考。
|
| 关 键 词： | 丘陵果园 农业机器人 单线激光雷达 树干检测 密度聚类 |
| 收稿时间： | 2020/3/3 0:00:00 |
Orchard Trunk Detection Algorithm for Agricultural Robot Based on Laser Radar |
| |
| NIU Runxin,ZHANG Xiangyang,WANG Jie,ZHU Hui,HUANG Jian,CHEN Zhengwei.Orchard Trunk Detection Algorithm for Agricultural Robot Based on Laser Radar[J].Transactions of the Chinese Society of Agricultural Machinery,2020,51(11):21-27. |
| |
| Authors: | NIU Runxin ZHANG Xiangyang WANG Jie ZHU Hui HUANG Jian CHEN Zhengwei |
| |
| Abstract: | In the view of the influence of slopes and weeds in orchard on the detection accuracy of fruit trees in hilly areas, a tree trunk detection algorithm based on adaptive density clustering was proposed. Firstly, the single line LiDAR was used to obtain the environmental information. That was through data preprocessing, the noise points and the unusable data points were filtered out, the clustering radius was set with the trunk as the target, and the clustering threshold was set adaptively according to the distance from the data points to the LiDAR, and then the preliminary clustering was completed. As following, the features of the preliminary clustering results and the data points in the ground class that huge also roughly in a straight line were used. After this, the class which was over the certain number of data point were used in the second curve fitting. Also, the class that fitting radius was greater than a certain threshold value was regarded as ground interference and needed to be eliminated. Finally, the class which data points were more than a certain number of adjacent data points were regarded as weed branches and leaves and eliminated by using the feature of discontinuous distance between data points in weed branches and leaves, thus the detection of tree trunks or orchard was completed. The experimental results showed that with no interference, the false detection rate was 0.76%, the missed detection rate was 1.90%, and the average accuracy rate was 97.3%, respectively;the average accuracy rate of tree detection was 96.1% when there was only ground interference;the average accuracy rate of tree detection was 91.4% when there was only weed interference, and the average accuracy rate of tree detection was 91.9% when there was both ground and weed interference. The overall average accuracy from all situations was 95.5%. This method could be used to detect trunk in arborization orchard with obvious tree trunk in hilly area and provide environmental understanding for the navigation of precision agricultural equipment in the orchard in hilly area. |
| |
| Keywords: | hilly orchard agricultural robot single line LiDAR trunk detection density clustering |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
|
