基于动态规划的拉班舞谱自动生成研究

发布时间：2018-05-13 23:30

本文选题：拉班舞谱 + 运动捕捉　；参考：《北京交通大学》2017年硕士论文

【摘要】：拉班舞谱是一种公认科学的动作分析和记录体系,是现在最广泛使用的动作谱,已经可以达到五线谱之于音乐的作用,常用于不同舞蹈艺术的交流。但是记谱仍然需要专业人士人工识别记录,十分耗时。运动捕捉是起源于20世纪动画技术的概念,现代运动捕捉技术已经相当成熟,广泛应用于电影、动画特效,可以达到十分逼真的效果,捕捉数据也日趋精确。但是设备成本居高不下。本次研究希望利用计算机识别动作,输出拉班舞谱,将运动捕捉设备作为眼睛,分析捕捉数据,获得动作识别结果,从而提高拉班记谱效率,为我国民族民间动态艺术的保护提供一种途径。本文介绍了一种基于动态规划的拉班舞谱自动生成方法,通过分析BVH(Bio-vision Hierarchical)格式的运动捕捉数据,识别各个基本动作,转换成舞谱,实现了舞谱生成的平台。首先通过一种被动式光学运动捕捉系统采集人体运动数据,保存为BVH文件。分析数据定义的骨骼层次结构,将节点与人体关节语义对应起来。然后将BVH格式的数据转换成易用的位置坐标数据以便后续分析。然后进行元素动作的分割分析,创新点在于在运动学分割的基础上,加入了节拍信息,从而对分割结果做时间上的规整,减小了分割误差,从而降低了分割不准对动作分析识别的影响。动作识别的准备上,由自主采集的数据对各个人体部位构建了元素动作模板库和元素动作样本库,最后采用动态时间规整的方法比较样本和已知动作类别的模板,最后得知样本动作类别,并在样本库上进行了识别正确率测试。拉班舞谱的输出上,采用一种能够对应拉班舞谱符号的数据结构,将符号与运动捕捉数据序列对应起来,从而能够将数据转换成舞谱。综上,本文研究完成了一种由运动捕捉数据借助计算机分析识别自动生成拉班舞谱的系统,并基于Python语言开发实现了系统平台,生成的舞谱与专家给出的标准舞谱对比,能够对目标的基本动作输出正确。
[Abstract]:Laban dance spectrum is recognized as a scientific action analysis and recording system. It is the most widely used action spectrum. It can achieve the function of music. It is often used in the exchange of different dance arts. But notation still requires professionals to manually identify records, which are time-consuming. Motion capture is a concept originated in the 20th century animation technology. Modern motion capture technology has been quite mature, widely used in movies, animation effects, can achieve very realistic effects, capture data is increasingly accurate. But equipment costs are high. The purpose of this study is to use the computer to identify the movement, output the Laban dance spectrum, take the motion capture device as the eye, analyze and capture the data, obtain the result of motion recognition, and improve the efficiency of Laban recording. It provides a way for the protection of national folk dynamic art in our country. In this paper, a method of automatic generation of Laban dance spectrum based on dynamic programming is introduced. By analyzing the motion capture data of BVH(Bio-vision hierarchy format, every basic motion is recognized and converted into dance spectrum, and the platform of generating dance spectrum is realized. Firstly, a passive optical motion capture system is used to collect human motion data, which is saved as BVH file. The skeleton hierarchy defined by the data is analyzed to correspond the joint semantics of the node and the human body. The data in BVH format is then converted into easy-to-use position coordinate data for subsequent analysis. Then the element action segmentation analysis, the innovation is based on the kinematics segmentation, adding the beat information, thus the segmentation results to make the time regular, reduce the segmentation error, Thus, the influence of segmentation inaccuracy on motion analysis and recognition is reduced. In the preparation of action recognition, the element action template database and the element action sample database are constructed from the self-collected data. Finally, the dynamic time regularization method is used to compare the template of the sample with the known action category. Finally, the category of sample action is known, and the correct recognition rate is tested on the sample database. In the output of Laban dance spectrum, a kind of data structure which can correspond to Laban dance spectrum symbol is adopted, and the symbol and motion capture data sequence are corresponding, thus the data can be converted into dance spectrum. To sum up, this paper studies and completes a system of automatically generating Laban dance spectrum from motion capture data by means of computer analysis and identification. The system platform is developed based on Python language, and the generated dance spectrum is compared with the standard dance spectrum given by experts. Able to output correctly to the target's basic action.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：J721;TP391.41

【参考文献】