Authors

Abstract

To address the Autonomous navigation and operation requirements of agricultural robots in complex terrain environments of vineyards, this report proposes a point cloud Ground segmentation algorithm based on surface fitting, aiming to solve the problems of reduced segmentation accuracy and insufficient adaptability of traditional planar assumption methods in unstructured terrains such as sloped fields and ridge furrows. The core of the algorithm lies in adopting a point cloud representation method based on a non-uniform polar grid, which dynamically allocates face element sizes according to point cloud density and the width between vineyard ridges, effectively addressing the issues of point cloud sparsity and representability. Subsequently, the moving least square method is used to fit surface models. During the fitting process, strategies such as Gaussian weight function, cosine and sine basis functions, and set of orthogonal functions are introduced to shorten the algorithm’s running time and reduce computational complexity. The algorithm’s performance is evaluated on the public dataset KITTI and in real-world environments, and compared with algorithms such as RANSAC, GPF, R-GPF, and Patchwork. Experimental results show that the proposed algorithm outperforms other algorithms in both Precision and Recall. In practical environments, the algorithm can accurately and effectively segment complex vineyard environments, meeting the operational requirements of agricultural robots and providing technical support for the advancement of smart agriculture.

Abstract in Chinese

针对葡萄园复杂地形环境下农业机器人自主导航与作业的需求，本文提出了一种基于曲面拟合的点云地面分割算法，旨在解决传统平面假设方法在坡地、垄沟等非结构化地形中存在的分割精度下降与适应性不足的问题。算法的核心在于采用一种基于非均匀极坐标网格的点云表示方法，该方法根据点云密度以及葡萄园垄间宽度动态分配面元大小，有效解决了点云稀疏性和可表示性问题；随后用移动最小二乘法拟合曲面模型，在拟合过程引入高斯型权函数、正余弦基函数，正交函数集的策略，缩短算法运行时间，降低计算复杂度。在公开数据集KITTI、葡萄园环境中评估算法性能，并于RANSAC、GPF、R-GPF、Patchwork等算法进行对比。实验结果表明，本文算法在Precision和Recall均优于其他算法；葡萄园环境中，该算法能准确、有效的分割葡萄园复杂环境，满足农业机器人的作业需求，为推进智慧农业提供了技术支撑。