基于株型参数虚拟量测及过程模型参数智能提取的植物三维重建方法

Models for accurately simulating plant morphologies and its growth mechanism not only can represent natural scenes, but also provide an effective and economic tool to optimize crop ideotype breeding and cultivation methods. However, the methods of obtaining the parameters of plant type and growth rules during the modeling process of virtual plant in current methods are still very inefficient and limited. Therefore, we are going to study automatic methods for measuring the topological and geometrical parameters of target plant based on the virtual model and intelligent methods to acquire the special morphogenetic model's rules and parameters, which are key techniques to reconstruct special 3D plant. This research includes: 1) analyzing the methods of obtaining target plant shapes based on its visible light images and depth images and the methods to reconstruct three-dimensional plant model; 2) based on the established virtual plant model, studying the principle and methods of automatically measuring the parameters of a plant type. 3) proposing a framework to describe the topological structure of plant based on the rules of a procedural model and studying effective strategies to put topological feature data of a target plant into the description framework of the procedural model, and presenting an intelligent method to acquire production rules to simulate the shapes of a target plant; 4) proposing methods to visually simulate the shapes of single plant and population of plant, which are based on hardware acceleration technology. The innovations in the project are: 1) the implicit mesh surface reconstruction method for plants based on depth data fusion; 2) the extraction algorithm of the parameters of skeleton shapes and forms of plants based on virtual plant model and pattern recognition methods; 3) intelligent model rule acquisition algorithm based on topology feature data and mapping method of predefined description framework. The key scientific problems are: how to improve the accuracy of our reconstruction algorithm, how to automatically extract parameters of a plant type based on its visual model, and how to intelligently get the regulation parameters of a procedural model. We have made some achievements in the fields of mesh reconstruction, plant modeling, plant growth mechanism and etc. Our studies will solve the difficult problems to acquire the parameters of a plant type and its growth rules during the process of modeling virtual plants, and can provide new methods and supporting techniques for agricultural applications, such as the dynamic evolution law of stems and leaves of a plant.

精确模拟植物形态及生长机理的仿真模型，可为自然景观再现、作物株型培育、栽培措施优化等提供有效工具。针对虚拟作物建模中株型参数及生长规则获取的低效性和盲目性的问题，项目拟结合植物三维重建，研究植物形态参数获取新方法和模型规则获取新方法，具体包括：基于图像的植物形状获取与三维重建；基于虚拟模型的株型参数测量方法；基于规则的植物拓扑描述框架，植物拓扑与描述框架的映射，形成模型规则获取方法。项目的创新在于：基于深度数据融合的植物隐式网格曲面重建方法、基于虚拟模型和模式识别方法的植物骨架形状与形态参数抽取算法、基于拓扑数据与描述框架映射的模型规则参数获取算法。关键科学问题为：如何提高重建算法的精度、如何依据模型自动提取株型参数以及如何智能化获取过程模型的规则参数。项目组在三维重建、植物建模、植物生长机理等方面有较强积累。项目有助于解决形态参数测量与规则获取的若干难点问题,也为植物演变研究提供新方法。

精确模拟植物形态及生长机理的仿真模型，可为自然景观再现、作物株型培育、栽培措施 优化等提供有效工具。针对虚拟作物建模中株型参数及生长规则获取的低效性和盲目性的问题，项目围绕着基于图像的植物形状获取与三维重建、基于虚拟模型的株型参数测量方法、基于规则的植物拓扑描述框架、植物拓扑与描述框架的映射、形成模型规则获取方法等内容开展了细致而又深入研究。通过四年的探索，联合申报单位通力合作，在以下多方面取得重要结果：提出了基于联合优化的精确深度信息估计方法和多视角深度信息融合与网格快速重建方法；通过对摄像头和Kinect传感器的标定，构建了基于深度传感器的远距离跟踪注册技术；提出了一种基于空间竞争、光照引导以及部分生物自然生长的规则的树木过程式建模方法和一种设计树木枝干层次细节模型的方法，可以有效的建立单棵或者多棵树在相互影响竞争下生长的模型以及仿真生长过程；为判断重建出的模型和真实植物间的相似性，提出了一种基于图像视觉特征的植物形态相似度计算方法；我们研究了基于重建的3D植物模型的株型拓扑结构参数和器官几何参数的自动提取方法。针对目前过程模型方法中描述方式不够直观、对输入要求严格以及耗时相对较长等问题，提出一种基于改进粒子群算法的虚拟植物生长模型参数提取方法，避免了人工提取参数的低效性和盲目性的问题；在研究的过程中，为解决粒子群优化算法易陷入局部最优值的问题，还提出一种新的引入多级扰动的混合型粒子群优化算法；研究了一种基于干物质流的西红柿生长可视化算法，提出了基于混合包围体树的碰撞检测算法，实现了并行化的植株群体碰撞检测及植株内部器官的碰撞检测，并在较大规模的番茄植株群体上进行了实验；最后，开发了植物拓扑结构参数智能获取及株型重建系统、基于PC平台的远距离增强现实系统。上述研究成果已通过试验得到了有效的验证。通过本项目的实施，发表了学术论文20篇（其中被SCI收录5篇，EI收录5篇），获得发明专利登记证书1件，申报发明专利3件，获得软件著作权2件。在研究生培养方面，培养博士研究生3人，硕士研究生5人。本课题负责人及项目组成员参加国际国内学术会议十余次。本课题组所取得的研究成果，已超额完成任务书有关技术水平的要求以及各项具体指标的要求。本研究在植物建模、株型参数及生长规则获取的理论和方法等方面，取得了一系列创新性研究成果，具有重要理论意义和较为广阔的应用前景。