基于Period的三维人体动作识别研究  

Research on 3D Human Action Recognition Based on Period

作　　者：苏文瑛[1] 刘艳[1] 李文博[1] 

机构地区：[1]天津大学计算机科学与技术学院,天津300072

出　　处：《计算机与现代化》2013年第4期90-94,共5页Computer and Modernization

摘　　要：在剖析现有人体运动建模方法的基础上,提出基于Period的建模方法。首先对关节点的运动规律做出基础假设,即每个关节点的相对运动都被限制在一个半球面上,利用半球模型来表征人体运动数据。然后提出Period的概念,并通过对节点运动路径的近似拟合,以多组Period序列来表示人体运动,使得数据规模大幅减小,最终实现对人体动作的匹配识别。实验表明,该方法能有效提高动作识别的精度和效率。Based on investigation of human motion modeling technology, a Period based modeling method is proposed. In this method, an assumption of human joints motion patterns is introduced, which points out every joints relative motion is limited on a haft-sphere surface, then a half-sphere model is used to represent motion data. Then concept of Period is proposed, and in order to reduce data scale, the motion trajectory is approximated by multi group of Period sequence. At last, human action matching and recognition is reached. Experiments show that this method can raise the accuracy and efficiency of human action recognition.

关 键 词：动作识别 半球模型 Period序列 路径拟合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]