基于三维运动捕捉数据关键帧的动画合成研究

发布时间：2018-05-02 14:40

本文选题：运动捕捉 + 关键帧提取　；参考：《江苏大学》2017年硕士论文

【摘要】：随着科技的不断进步,动画合成技术的应用越来越广泛。人们对动画合成技术有了新的功能需求,那就是对现有的动作进行重用,从而产生一些不同的效果。即采用动画合成技术给用户一种强烈的视觉冲击,并带给他们震撼的喜剧效果。而现有的动画合成技术不能满足这种需求。针对该问题,本文作了相关的课题研究,旨在通过已有的动作生成一种与众不同的动画效果。本文首先从课题的研究背景和意义展开,深入研究课题的相关应用情况和现存问题,设计并实现了本文的研究方法:一种基于三维运动捕捉数据关键帧的动画合成方法。然后围绕课题的相关工作,从动画合成的技术内容进行阐述,简单介绍了运动合成和运动重定向在三维动画合成领域中的技术特征,进一步介绍了二者所需的关键技术,主要有关键帧提取技术和动作捕捉的分割技术。在已有的理论研究基础上,本文设计了和提出了两种关键帧提取方法。最后基于本文设计和提出的方法,利用插值的思想,设计并且实现了基于三维运动捕捉数据关键帧的动画合成系统。本文主要研究内容如下:(1)设计并实现了基于余弦距离层次聚类的运动捕捉数据关键帧提取方法。该方法充分利用关节点的旋转量来作为分割运动捕捉数据的特征值,接着去除捕捉数据中的噪音,然后通过降维的方法将高维数据映射为低维数据。随后用余弦距离计算相似度,紧接着采用层次聚类分割,将各个分割点和各段中的帧姿势与均值误差最小的帧作为关键帧序列。实验表明,相比较基于速率的分割方法和基于曲率的分割方法,该方法具有较高的准确率和查全率。(2)提出了基于最优分割的运动捕捉数据关键帧提取方法。考虑到基于余弦距离层次聚类的运动捕捉数据关键帧提取方法仅是通过经验启发式设计的,我们对关键帧的求解过程做了进一步的扩展。该方法首先引入模型机制建立模型,运用该模型将运动捕捉数据的分割问题建模成一个有序样本聚类问题。其次,采用最优分割算法对运动捕捉数据进行分段,求出最小化段内平方误差和。最后提取出分割点和每段中帧姿态与均值误差最小的帧作为关键帧。实验结果表明,与DWT以及PAA算法进行相比,所提出的方法具有较好的可视化结果,所得关键帧在表达原始运动捕捉数据上具有一定优势,能够对原始运动捕捉数据进行较好的概括和总结。(3)设计并实现了基于三维运动捕捉数据关键帧的动画合成系统。基于以上的两种方法,结合贝塞尔曲线算法,运用软件设计过程中面向对象的思想,设计并实现了基于三维运动捕捉数据关键帧的动画合成系统。该系统不仅界面友好,操作简单,功能齐全,而且系统的动画片段重用性高,能够实现别具一格的动画效果,进一步验证了方法的可用性。
[Abstract]:With the development of science and technology, the application of animation synthesis technology is more and more extensive. Animation synthesis technology has a new functional requirement, that is, to reuse the existing actions, thus producing some different effects. That is, using animation synthesis technology to give users a strong visual impact, and bring them shocking comedy effect. The existing animation synthesis technology can not meet this demand. To solve this problem, this paper makes a related research, aiming to create a distinctive animation effect through existing actions. In this paper, the research background and significance of the subject are discussed, and the related applications and existing problems are deeply studied. The research method of this paper is designed and implemented: an animation synthesis method based on the key frames of 3D motion capture data. Then around the related work of the subject, the technical content of animation synthesis is expounded, and the technical characteristics of motion synthesis and motion redirection in 3D animation synthesis field are briefly introduced, and the key technologies required by them are further introduced. There are key frame extraction techniques and motion capture segmentation techniques. Based on the existing theoretical research, this paper designs and proposes two key frame extraction methods. Finally, an animation synthesis system based on the key frames of 3D motion capture data is designed and implemented based on the method designed and proposed in this paper and the idea of interpolation. The main contents of this paper are as follows: 1) the key frame extraction method of motion capture data based on cosine distance hierarchical clustering is designed and implemented. This method makes full use of the rotation of the node as the eigenvalue of the segmentation motion capture data, then removes the noise from the captured data, and then maps the high-dimensional data to the low-dimensional data by reducing the dimension. Then the similarity is calculated by cosine distance, and then hierarchical clustering is used to segment the frame pose and the frame with the minimum mean error in each segmentation point and segment as the key frame sequence. Experimental results show that compared with rate-based segmentation and curvature based segmentation, the proposed method has high accuracy and recall rate. (2) A motion capture key frame extraction method based on optimal segmentation is proposed. Considering that the key frame extraction method of motion capture data based on cosine distance hierarchical clustering is only designed by empirical heuristic we extend the key frame solving process. Firstly, the model mechanism is introduced to model the segmentation of motion capture data as an ordered sample clustering problem. Secondly, the optimal segmentation algorithm is used to segment the motion capture data to minimize the sum of square error. Finally, the frame with the minimum error between the attitude and the mean value of each segment is extracted as the key frame. The experimental results show that the proposed method has better visualization results than DWT and PAA algorithms, and the key frames have some advantages in representing the original motion capture data. The animation synthesis system based on the key frame of 3D motion capture data is designed and implemented. Based on the above two methods and the algorithm of Bezier curve, an animation synthesis system based on the key frame of 3D motion capture data is designed and implemented by using the object-oriented idea in the software design process. The system not only has friendly interface, simple operation, complete function, but also has high reusability of animation fragment, which can realize unique animation effect, and further verify the usability of the method.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】