步态驱动的四足动物低维物理运动生成研究

发布时间：2018-08-30 20:18

【摘要】：角色动画是计算机动画领域的核心问题之一,其研究目标在于依据动画角色的自身属性、所处环境、运动目标等约束,快速有效地生成相应的运动序列。两足和四足角色是角色动画领域的两个主要研究对象。其中,随着运动捕捉设备的广泛使用和两足角色运动数据的大量积累,两足角色运动合成与识别研究得以迅速发展。然而,以四足陆地动物为代表的四足角色形态多变、种类繁多,适合运动捕捉的物种却屈指可数。造成了四足动物的运动数据稀缺,进而导致了四足角色运动编辑与合成的研究相对滞后。另一方面,相对于两足角色,四足角色在运动时拥有更为丰富的步态模式和足迹约束,并且结构平衡性更为稳固,使得基于物理模型的运动生成方法成为获取四足动物运动序列的有效渠道。因此,本文使用低维物理模型对四足动物运动模型进行抽象,并以动态的步态约束驱动低维物理运动生成,从而提出了一种面向四足动物的运动生成方法。用户仅需要输入直观简便的步态图或速度信息,辅以少量的交互代价即可快速生成自定义四足运动。本文的主要研究内容包括:(1) 一种面向四足动物运动生成的低维物理模型。提出了一种面向四足运动的低维物理模型:首先,该模型使用“质点”和“刚体约束”来描述骨骼动画中动画角色的“关节点”和“骨骼”,并建立了相应的物理解算器;其次,通过建模线弹簧和扭力弹簧结构,对躯干、四肢、头尾等肢体分别进行低维抽象;最后,基于四足动物生理和物理自由度的先验知识,构造约束队列修正低维物理模型的生成结果。基于该模型,用户通过少量的交互即可快速生成具有视觉真实感的四足动物运动,也能够通过参数微调生成具备自定义风格的运动。(2) 一种面向低维物理运动生成的动态步态约束模型。提出了一种动态步态约束模型。该模型能够依据瞬时速度动态地对步态模式和步态参数进行规划,也能够通过识别步态图估测运动速度。而后,基于所得到的步态参数和速度,并依据生物在运动中所遵守的能量优化原则,建立腿部关节运动与时间、地形、步态之间的关联模型,实时规划每次摆腿动作的自适应足迹曲线。对于运动转向的情形,制定了头部、躯干和足迹各自的转向策略,从而能够依据目标方向生成相应的足迹和全身约束。(3) 一个通用于四足动物的快速运动生成系统。为了辅助用户快速生成多样化的运动,在现有步态约束模型和低维物理模型基础上,为提高数据和参数的可重用性进行了相应的研究,其主要内容包含:针对输入的四足骨架的自动关节匹配和语义识别、基于形态相似度的层次化运动重定向以及面向风格化运动生成的语义标签运算规则。在这些基础上,将研究成果整合为一套能够自动批量导入角色、快速生成目标运动、并支持通过语义标签操作实现风格化的四足动物运动生成的系统。
[Abstract]:Character animation is one of the core issues in the field of computer animation. Its research goal is to generate corresponding motion sequences quickly and effectively according to the character's own attributes, the environment and the constraints of moving objects. With the extensive use of biped character motion data and the accumulation of biped character motion data, the research on biped character motion synthesis and recognition has been developed rapidly. However, the quadruped terrestrial animals represented by quadruped terrestrial animals have a variety of shapes and species, but few species are suitable for motion capture. On the other hand, compared with biped characters, quadruped characters have more abundant gait patterns and footprint constraints, and their structural balance is more stable, which makes the method of motion generation based on physical model become an effective way to obtain quadruped motion sequences. A quadruped-oriented motion generation method is proposed by abstracting the quadruped motion model from the low-dimensional physical model and driving the low-dimensional physical motion generation with dynamic gait constraints. Users need only to input intuitive and convenient gait maps or speed information, and with a small amount of interaction cost, they can quickly generate customizations. The main research contents of this paper include: (1) A low-dimensional physical model for quadruped motion generation. A low-dimensional physical model for quadruped motion is proposed. Firstly, the model uses "particle" and "rigid constraint" to describe the "joint" and "bone" of animated characters in skeletal animation, and establishes the model. Secondly, low-dimensional abstraction of trunk, limbs, head and tail is carried out by modeling the structure of line spring and torsion spring. Finally, based on the priori knowledge of quadruped physiology and physical freedom, a constraint queue correction low-dimensional physical model is constructed. A dynamic gait constraint model for low-dimensional physical motion generation is proposed. The model can dynamically generate gait patterns and gaits according to instantaneous velocity. Then, based on the obtained gait parameters and speed, and according to the energy optimization principle observed by organisms in the movement, the correlation model between leg joint motion and time, terrain and gait is established, and the adaptive footprint curve of each swing leg movement is planned in real time. (3) A general fast motion generation system for quadrupeds. In order to assist users to quickly generate diverse motions, existing gait constraint models and low-dimensional physics are used. On the basis of the model, the reusability of data and parameters is improved. The main contents include: automatic joint matching and semantic recognition for input quadruped skeleton, hierarchical motion redirection based on morphological similarity and semantic label operation rules for stylized motion generation. The research results are integrated into a set of quadruped motion generation system which can automatically import roles in batches, quickly generate target motion, and support stylized quadruped motion generation through semantic tagging operation.
【学位授予单位】：合肥工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】