【摘要】 为满足中小型农业机械在无作物或种植低矮作物的单栋塑料温室自主导航作业需求，提出一种基于钢骨架检测来实现导航参数获取的方法。对室内环境结构进行分析；通过欧氏变换、直通滤波和DBSCAN聚类，从三维激光点云数据中提取出距离最近的侧窗点云集合；为获取稳定的钢骨架点云，提出一种钢骨架检测方法，考虑其空间位置等因素，设置不同阈值滤除卷膜点云并提取出符合条件的钢骨架点云；采用主成分分析法拟合导航基准线并计算出导航参数。试验结果表明，本文方法对钢骨架检测有效率不低于88%;获取的横向距离和偏航角平均绝对误差分别为0.03 m和2.12°。本文方法能满足农业机械在单栋塑料温室中自主导航作业的需求，可为该环境中的自主导航研究提供参考。更多还原

【Abstract】 With the aim of meeting the autonomous navigation requirements of small and medium-sized agricultural machinery in tunnel greenhouse with no crops or low crops, a navigation line fitting method based on steel tube detection was proposed. Firstly, the indoor environment structure was analyzed. Secondly, the greenhouse point cloud data was collected by LiDAR, and the nearest side window point convergence was extracted from the 3D laser point cloud data by Euclidian transformation, pass-through filtering and DBSCAN clustering. Then a steel tube detection method was proposed with the intention of obtaining stable steel tube point cloud, which iterated through all the scanning lines of the target window point cloud, considering its spatial location and other factors, set different thresholds to filter out the plastic film point cloud and extract the qualified steel tube point cloud. Finally, a straight line fitted by the steel tube point cloud was obtained by principal component analysis, and the navigation parameters of the platform in the greenhouse were determined by using this as the navigation reference line. The steel tube detection test showed that the efficiency of the steel tube detection method in tunnel greenhouses was not less than 88%, and the navigation reference line fitted by the proposed algorithm had a high updating frequency. In the navigation parameter measurement accuracy test, the mean absolute errors of cross range and yaw angle were 0.03 m and 2.12°, respectively. The research result can meet the requirements of autonomous navigation of agricultural machinery in a tunnel greenhouse, which can provide reference for the research of autonomous navigation in this environment.更多还原