【作者】 罗冠；

【导师】 郝重阳；

【作者基本信息】 西北工业大学 ， 信号与信息处理， 2004， 博士

【摘要】 受虚拟现实技术飞速发展的驱动和客观应用需求的牵引,虚拟人技术逐渐成为虚拟现实技术的一个重要分支,已经越来越受到人们的重视。对这一前沿性问题的研究已形成新的研究热点。 虚拟人是人在计算机生成空间中几何特性与行为特性的逼真表示。虚拟人技术是研究适人化虚拟空间、高度逼真感与沉浸感虚拟环境的重要研究内容,它在为人们解决安全与健康、教育与娱乐以及产品设计等许多重大实际问题时,提供了新方法和新途径。 虚拟人的运动生成及控制是虚拟人技术的重要研究方向。本论文对虚拟人的骨骼结构表达、肢体的运动生成及控制等一系列问题进行了系统、深入地研究,取得的创新性成果主要有: 1.提出了一种能够广泛表现人类基本运动的最小人体模型。论文通过合理地简化与假设,只保留了最能体现人类运动低频特性的关键关节和运动自由度,提出了一种16关节、39自由度的虚拟人简化模型。仿真结果表明,该模型在保证虚拟人运动逼真性的前提下,不仅有效降低了研究工作的复杂性、提高了虚拟人运动仿真的实时性,而且使得在PC平台上研究虚拟人的运动生成及控制技术成为可能。 2.提出了一种基于多刚体系统模型的虚拟人骨骼结构的建模方法。论文首先将虚拟人抽象成为一个各肢体间通过机械转动铰相连的多刚体系统,然后引入树模型与低序体阵列来表达虚拟人的骨骼结构。与传统的DH表示方法相比,本文方法具备以下优点:(1).能够自然地表现虚拟人的复杂分支结构;(2).能够很容易地实现对最小人体模型的任意精度扩展。 3.提出了一种基于Bezier样条曲线的虚拟人运动轨迹的构造方法。论文采用Bezier样条曲线构造人体重心及摆动腿脚部的运动轨迹。与基于三角函数拟合的方法相比,本文方法具有更强的复杂空间运动轨迹的表达能力及运动曲线的交互控制能力;同时也比Hermite样条曲线具备更直观的运动轨迹构造与调整能力。 4.提出了一种基于限时梯度投影的虚拟人行走运动的生成方法。论文将虚拟人的逆向运动学求解问题转变为一个约束非线性规划问题,通过动态调整每帧的规划时间,可同时达到非线性函数局部线性化和限时求解的双重目的。与传统Jacobian矩阵广义逆的求解方法相比,本文方法可避免求解过程中的多值问题与奇异性问题;与变尺度法和共轭梯度法相比,本文方法可保证线性收敛,在PC平台上可基本达到实时效果。 5.提出了一种虚拟人的广义行走模型及基于交互控制的行走运动生成方法。为表现虚拟人行走运动的多样性与复杂性,论文首先提出了虚拟人广义行走模型的概念;然后通过对虚拟人规范行走模型进行交互控制与调整,可以实时生成形态各异的虚拟人广义行走运动。与运动捕捉方法相比,本文方法具有廉价、直观的优点,能够有效提高行走模型的可重用性。更多还原

【Abstract】 With the in-depth development of virtual reality technology and the increasing requirement for it, the modeling and simulation of virtual human in the virtual environment have attracted more and more attention, gradually forming a new hot research field.Virtual human, as the vivid expression of human’s geometrical feature and behavior characteristic in computer generated space, is the essential topic for the research of human-fit virtual space and highly immersive virtual environment. It provides a new way for human when dealing with crucial problems, such as safety and healthiness, education and entertainment, product design, etc al.The motion generation and control techniques are one of the most significant aspects of virtual human technology. The representation of the virtual human’s skeleton structure, the motion control of the limbs and trunk, and a series of related problems are systemically and deeply studied in the thesis. The main work and contributions are summarized as follows:1. A minimal human model is presented which can be widely used to express the virtual human’s basic movement. After thorough study of the skeleton structure of human body, a simplified virtual human model with 16 joints & 39 degrees of freedom is presented in the thesis. This is based on reasonable simplification and assumption, and only those key joints and degrees of freedom which can mostly represent the low-frequency characteristic of human motion are kept. The simulation results show that our model could help to decrease the complexity of the research work, increase the real-time performance, and make it possible for studying the motion generation and control techniques on PC platform without much loss of fidelity.2. A skeleton modeling method based on multi-rigid-body system theory is proposed in the thesis. Firstly virtual human is abstracted as a multi-rigid-body system whose limbs and trunk are coupled with mechanical revolute joints. Then tree model and lower body array are introduced to represent the skeleton structure. Compared with traditional DH method, our model could naturally exhibit the complex branched characteristic of virtual human and could easily realized arbitrary resolution extension for the minimal human model.3. A motion trajectory constructing method based on Bezier spline curve is put forward. Bezier spline curve is adopted to form the motion track of the body center of mass and the foot of swing leg. Compared withtrigonometric function approximation algorithm, our method has stronger representation and interactive control capability for complex space curves. It is also much more intuitive for motion curve construction and adjustment than Hermite spline curve.4. A walking model for virtual human based on time-critical gradient projection method is presented. Unlike the traditional Jacobian matrix method, we first convert the inverse kinematics problem to a constraint nonlinear programming problem. Then by changing the programming time of every frame, local linearization of the nonlinear function and time-critical computing can be achieved simultaneously. The multi-solution and singularity problems can be avoided because there is no need to compute the pseudoinverse any longer. Compared with variable metric method and conjugate gradient method, our algorithm could assure linear convergence, and near real-time performance could be obtained on PC.5. A generalized walking model and the interactive control method are presented. In order to show the diversity of complex motion for virtual human, a generalized walking model is proposed. Then four different kinds of motion based on the walking model is described and realized by interactively controlling and adjusting the regular walking model. Compared with motion capture method, our method is simpler and cheaper. It also promotes the reusability of the waling model.更多还原