Wen Weiliang, Li Baojun, Li Bao-Jun, Guo Xinyu
Beijing Research Center for Information Technology in Agriculture, Beijing, China.
Beijing Key Lab of Digital Plant, National Engineering Research Center for Information Technology in Agriculture, Beijing, China.
Front Plant Sci. 2018 Jun 26;9:783. doi: 10.3389/fpls.2018.00783. eCollection 2018.
This paper introduces a novel hierarchical structured representation for leaf modeling and proposes a corresponding multi-resolution remeshing method for large-scale visual computation. Leaf modeling is a very difficult and challenging problem due to the wide variations in the shape and structures among different species of plants. Firstly, we introduce a Hierarchical Parametric Veins and Margin (HPVM) representation approach, which describes the leaf biological structures and exact geometry via interpolation of parametric curves from the extracted vein features from non-manifold data. Secondly, a parametric surface model is constructed using HPVM with geometric and structured constraints. Finally, for a given size, we adapt a multi-step discrete point resampling strategy and a CDT-based (Constrained Delaunay Triangulation) meshing method to generate a mesh model. Our representation consists of three coupled data structures, a core hierarchical parametric data structure of veins and margin for the leaf skeleton, the corresponding parametric surface model, and a set of unstructured triangular meshes with user-specified density for the leaf membrane. Numerical experiments show that our method can obtain high quality meshes from the scanned non-manifold mesh data with well-preserved biological structures and geometry. This novel approach is suitable for effective leaf simulation, rendering, texture mapping, and simulation of light distribution in crop canopies.
本文介绍了一种用于叶片建模的新型层次结构表示方法，并提出了一种相应的多分辨率重网格化方法，用于大规模视觉计算。由于不同植物物种的形状和结构差异很大，叶片建模是一个非常困难且具有挑战性的问题。首先，我们引入了一种层次参数化叶脉和叶缘（HPVM）表示方法，该方法通过对从非流形数据中提取的叶脉特征进行参数曲线插值来描述叶片的生物结构和精确几何形状。其次，利用HPVM并结合几何和结构约束构建参数化曲面模型。最后，对于给定的尺寸，我们采用多步离散点重采样策略和基于约束德劳内三角剖分（CDT）的网格化方法来生成网格模型。我们的表示由三个耦合的数据结构组成，一个用于叶片骨架的叶脉和叶缘的核心层次参数数据结构、相应的参数化曲面模型以及一组具有用户指定密度的用于叶片表面的非结构化三角形网格。数值实验表明，我们的方法能够从扫描的非流形网格数据中获得高质量的网格，同时很好地保留生物结构和几何形状。这种新颖的方法适用于有效的叶片模拟、渲染、纹理映射以及作物冠层光分布的模拟。
