摘要
作业场景重建可为智能农机自主作业提供全局信息与局部细节,针对因农田表面缺乏高区分度的点、线、面高层结构造成的特征描述性差、点云配准精度不足的问题,提出一种基于旋转曲面轮廓特征的农田地表点云配准方法。首先,采用32线激光雷达获取农田真实地表点云数据并完成去噪、降采样等预处理;然后,采用加权线性协方差矩阵的奇异值分解确定关键点唯一局部参考坐标系,并统计关键点与旋转曲面截面交点距离信息,生成地表点云的局部特征;最后,采用基于单特征初选与局部特征精匹配原则的多级特征匹配策略进行局部特征匹配,计算旋转矩阵与平移矩阵完成点云配准。试验结果表明,旋转曲面轮廓特征与其他特征相比,平均精度增加7.5个百分点,平均召回率增加24.09个百分点;多级特征匹配策略相对于最近邻搜索策略,平均精度增加12.68个百分点,平均召回率增加18.38个百分点;本文的点云配准方法的平均平移误差为23.59dr,平均旋转误差为3.72°,配准成功率为87.5%。因此,本文提出的基于旋转曲面轮廓特征的农田地表点云配准方法适用于真实农业地表无序点云的自动配准。
Scene reconstruction can provide global information and local details for the autonomous operation of intelligent agricultural machinery.Aiming at the problem of poor feature description and insufficient point cloud registration accuracy caused by the lack of high-level structure of points,lines and planes on the surface of field,a solution based on point cloud registration method for farmland surface based on contour features of rotating surface was proposed.Firstly,the 32-line LiDAR was used to obtain real surface point cloud data of the field and complete pre-processing such as denoising and down-sampling;then,singular value decomposition of weighted linear covariance calculation matrix was used to determine the unique local reference coordinate system of key points,and the distance information of the intersection between the key points and the rotating surface section was calculated to generate the local feature descriptor of the surface point cloud;finally,a multi-level feature matching strategy based on the principle of single feature primary selection and local feature fine matching was used to perform local feature matching,and the rotation matrix and translation matrix were calculated to complete the point cloud registration.The analysis results showed that compared with other methods,the average accuracy of the contour feature of the rotating surface was increased by 7.5 percentage points,and the average recall rate was increased by 24.09 percentage points;compared with the nearest neighbor search,the multi-level feature matching strategy increased the average accuracy by 12.68 percentage points and the average recall rate by 18.38 percentage points;the point cloud registration method proposed had an average translation error of 23.59dr,an average translation error of 3.72°,and a registration success rate of 87.5%.Therefore,the proposed field surface point cloud registration method based on the contour feature of the rotating surface was suitable for the automatic registration of the real agricultural
作者
董乃希
迟瑞娟
杜岳峰
温昌凯
张真
DONG Naixi;CHI Ruijuan;DU Yuefeng;WEN Changkai;ZHANG Zhen(College of Engineering,China Agricultural University,Beijing 100083,China;Beijing Key Laboratory of Optimized Design for Modern Agricultural Equipment,China Agricultural University,Beijing 100083,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2020年第S01期325-332,377,共9页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2016YFD0701901)。
关键词
作业场景重建
农田地表特征
点云配准
旋转曲面轮廓特征
多级特征匹配策略
work scene reconstruction
field surface characteristics
point cloud registration
rotating surface profile features
multi-level feature matching strategy
