【作者】 程熙；

【导师】 潘志庚；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2010， 博士

【摘要】 随着体育训练、军事作战、工程测试、计算机游戏以及影视娱乐等行业对人体动画需求的日益增加,人们不仅要求虚拟人的运动高度真实,而且希望虚拟人能灵活地与外界交互并具有智能。本文以生成交互式的真实感虚拟人运动为主要研究内容,分别对虚拟人运动的混合驱动方式、平衡和保护动作生成、用户控制的虚拟人运动、目标导向的运动合成等方面进行了较为深入的研究,提出了相应的算法和实现方法,具体研究内容包括：提出一种结合运动捕获数据驱动和动力学模拟的混合控制算法。本算法采用了两个并行计算进程,通过恢复运动数据集对神经网络进行训练,并根据虚拟人的恢复动作策略,借助该神经网络对运动数据库做预分类,有效地降低了后续的搜索范围。实验结果表明,本算法可以驱动虚拟人在两种运动控制方式之间实现灵活的切换。提出一种虚拟人运动中受扰的平衡保持和保护动作生成算法。本算法能快速检测虚拟人在运动中的平衡状态,并在虚拟人有摔倒倾向时驱动虚拟人采取自主的保护动作以减少倒地对身体造成的伤害。实验结果表明,本算法设计的平衡保持和保护动作控制器适合应用在交互式游戏及其他实时人体动画生成中。提出一种基于运动片断的交互式虚拟人运动生成算法。本算法采用支持向量机预测用户未来的输入信号,并生成控制策略表,能快速选择最优的下一个运动片段并生成最终的交互式人体动画。实验结果表明,本算法可以获得较高真实感的交互式人体动画,并且控制策略生成的时间-空间开销均较低。提出一种目标导向的虚拟人运动片段优化搜索和控制算法。本算法由用户提供少量关于最终生成运动的信息,系统能根据场景特征自动设定障碍物约束,并基于一个合适规模的运动片断数据库,用优化方法不断地搜索出运动片断,最终生成尽可能满足用户要求且视觉上自然的人体动画。实验结果表明,本算法在虚拟体操运动的应用中能较好地生成满足用户指定要求的人体动画。更多还原

【Abstract】 With the increasing requirements of character animation in sport training, military training, engineering testing, computer games and movies, the generated character motions are not only required to be highly realistic, but the characters are also expected to intelligently interact with the environment. Generating realistic and interactive virtual character motion is the main research content of this dissertation, which focuses on the hybrid driven approach of the character motion, the balance keeping and protective actions, the user-controlled virtual character and the goal-oriented character motion synthesis. In this dissertation, we put forward the corresponding algorithms and their implementations, which including:We present a hybrid control algorithm which combines the motion capture data and the dynamic simulation techniques together. We employ a parallel algorithm of two processes, as well as training an artificial neural network by a set of recovery motion capture data sequences, and predicting and pre-classifying the recovery motion database according to the recovery motion strategy of the characters, so reducing the size of the search region. The experimental results indicate that our characters can be controlled and switched between motion capture and dynamic simulation control modes naturally.We present a character balance keeping and protective motion generation algorithm in locomotion. The algorithm can detect the balance of the character in locomotion and drives him to take several active and protective responses when he falls to the ground. The experimental results indicate that our balance keeping and protective motion controller is suitable for interactive games or other real-time applications.We present a motion fragment based interactive character motion generation algorithm. We employ a SVM to add the prediction of the user’s future inputs in the control process, then selecting a motion fragment, and finally creating a motion stream from several short motion fragments to generate interactive motion. The experimental results indicate that our control method achieves higher visual quality and lower space-time costs in the control policy synthesis than the previous methods.We present a goal-oriented character motion fragment optimization search and control algorithm. The user provides information of the final character motion, and the system sets the obstacles of the scene automatically. Based on a motion capture database with appropriate size, our algorithm is able to continually search out the motion fragments, so that the final motion has high visual effect, as well as meeting the user’s requirements as much as possible. The experimental results indicate that the algorithm is suitable for user-specified gymnastics animation synthesis.更多还原