基于SOM特性的三维运动检索研究

发布时间：2018-04-09 02:05

本文选题：人体运动捕捉数据库　切入点：SOM　出处：《大连大学》2011年硕士论文

【摘要】：运动捕获技术的逐步发展和设备技术的进步极大地促进了大量三维人体运动捕捉数据的形成,其应用领域也广泛的拓展到到计算机动画、电影特技等。因此,目前图形学领域和动漫等应用领域的研究热点转向了对于人体运动捕捉数据的研究。然而,在对运动捕捉数据进行各种处理之前,必须能够先从已有的人体运动捕捉数据库中检索出需要的运动,而且要迅速、准确。因此,如何有效的利用计算机技术,能够自动地从运动捕捉数据库中迅速而又准确地检索出需要的各种运动是一个亟待解决的问题。本文是在利用自组织映射神经网络(SOM)的拓扑特性对人体运动捕捉数据进行特征映射的基础上,提出了基于SOM特性和主成分分析(PCA)索引相结合的三维运动检索,以及基于SOM特性和加权马氏距离相结合的三维运动检索两种算法。为了简化运动检索的流程,本文利用SOM的特征映射来实现特征提取和数据降维的结合,而普通的SOM必须经过拓扑特性加强处理才能来进行特征提取。利用SOM把每一种运动都映射到特征曲面之后,一种思路是利用PCA算法提取特征曲面的最大特征向量建立索引机制,加快检索的速率;一种思路是在特征曲面的基础上通过PCA提取主成分,然后利用主成分的贡献率确定加权马氏距离的权值,最后计算加权马氏距离进行相似性比较。本文对两种算法都进行了仿真比较,实验结果证明了两种算法的有效性。
[Abstract]:The gradual development of motion capture technology and the progress of equipment technology greatly promote the formation of a large number of three-dimensional human motion capture data, and its application fields are also widely extended to computer animation, film stunts and so on.Therefore, the current research focus in graphics and animation applications has turned to the study of human motion capture data.However, before processing the motion capture data, we must be able to retrieve the required motion from the existing human motion capture database, and must be quick and accurate.Therefore, how to effectively use computer technology to automatically retrieve all kinds of motion from the motion capture database quickly and accurately is an urgent problem to be solved.Based on the feature mapping of human motion capture data based on the topological characteristics of self-organizing mapping neural network (SOM), this paper proposes a 3D motion retrieval method based on the combination of SOM characteristics and principal component analysis (PCA) indexes.And three-dimensional motion retrieval algorithm based on SOM and weighted Markov distance.In order to simplify the process of motion retrieval, the feature mapping of SOM is used to realize the combination of feature extraction and data dimensionality reduction, while the common SOM must be strengthened by topological characteristics to extract features.After using SOM to map every motion to feature surface, one idea is to use PCA algorithm to extract the maximum feature vector of feature surface and build index mechanism to accelerate the speed of retrieval.One idea is to extract the principal component by PCA on the basis of the characteristic surface, then determine the weight of the weighted Markov distance by using the contribution rate of the principal component, and finally calculate the similarity of the weighted Markov distance.The two algorithms are simulated and compared in this paper. The experimental results show that the two algorithms are effective.
【学位授予单位】：大连大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TP391.41

【参考文献】