基于样本的植物模型合成及重建技术研究

Since plants are among most common elements that constitute the nature world, the three-dimensional models of plants are extensively used in nearly all types of computer graphics application such as animated movies and video games. These applications require large amount of detailed and realistic 3D plant models. In recent years, some techniques can generate 3D representations of plants from free hand sketching or images or scans. However, it is a cumbersome and time-consuming work to create each plant manually in a gigantic virtual environment. Other approaches use rule-systems or procedural modes to generate new plants from initial state. However, a certain professional knowledge is demanded for 3D modelers. This project explores an example-based modeling technique designed for a single plant and flora models. A small example model is provided first, and then the algorithm generates a large model with similar styles and structures that resemble the example model. The geometry of plants is determined by its branching structure, and other organs. And the models of plant contain rich visual information and sematic global context. Based on the observations above, the research involves geometric analysis for the example, branching structure synthesis, organ synthesis, interactive controlling and plant reconstruction from point clouds and examples. The research emphasizes on two aspects: a proper definition of local similarity between skeletal structures of plants in a continuous space; global constrains extracted from the example to guarantee the consistence between results and examples. This major contribution of the research is to develop an example-based algorithm that generates 3D models by resembling an input model, which could generate a variety of complex and rich environments simply and effectively.

目前电影游戏等应用对植物模型的需求量越来越大，而植物形态的复杂性和种类的多样性给植物建模方法带来了很大的挑战。目前的建模方法，利用事先构建植物模型库或结合植物生长规律，基于规则文法、草绘图或真实数据（图片或者点云）来构建植物模型。植物个体和植物群体都具有自相似性和生长性的典型特点，因此，单个植物模型样本具有丰富的信息，能合成类似风格和结构的多个模型。项目拟解决两个关键科学问题：量化植物结构之间的局部相似性、定义约束使合成结果与样本的保持全局一致性。本项目的研究内容包括：分析样本的枝干结构等几何信息，根据提取的枝干结构合成新的枝干结构，将细枝、叶花等植物器官添加到合成的枝干上来构建完整模型，定义合成空间的形状和特征等交互式控制方式，基于真实数据和样本重建三维植物模型。本项目旨在研究基于样本的植物建模技术、用于从样本中合成或重建相同风格和结构的模型，为种类繁杂的复杂植物场景构建提供技术基础。

围绕着基于样本的植物模型合成技术研究，从精细植物样本的获取和重建、骨架结构的分析与合成、植物器官（花朵）的开放过程三个方面展开了系统性的研究，取得了一系列创新性成果，相关成果发表在图形学领域国际会议和期刊（Eurographics, Computer Graphics Forum）上，申请了多项发明专利和软件著作权，顺利完成了预期目标。