基于序列比对的简单动作识别  

Simple movement recognition based on sequence alignment

作　　者：安香子[1] 刘冀伟[1] 么键[1] 

机构地区：[1]北京科技大学信息工程学院,北京市100083

出　　处：《中国组织工程研究与临床康复》2007年第40期8062-8065,共4页Journal of Clinical Rehabilitative Tissue Engineering Research

摘　　要：研究一种新的简单动作的识别方法,以达到在提高行为识别准确度的同时,又能有效降低计算复杂度的效果。①方法介绍:首先对进行运动分割及预处理后的图像序列建立完整的光流场,即通过计算光流方程的Hessian矩阵,把Hessian矩阵条件数的倒数作为Lucas-Kanade光流法的加权阵,这样可有效消除局部邻域中不可靠约束点,同时提高基本约束方程解的稳定性。然后借用灰度直方图的概念构造方向直方图,提取光流特征,方向直方图由量化光流场来得到,再利用方向直方图构造出一个模板库及其索引序列库。最后通过序列比对的方式来计算不同运动序列所表示的行为之间的相似度,当在线序列与参考序列的最大相似度大于一个阈值时,行为识别模块就认为当前序列是参考序列所对应的行为。②对人体的几个不同的简单动作进行了试验,证明了上述方法的有效性。In order to improve the accuracy of behavior recognition and effectively reduce the complexity of computing, this paper is aimed to introduce a new method to recognize simple movements.①First an integrity optical flow field of the image sequences following motion segment and pre-procession is built. That means to calculate the Hessian matrix of optical flow equation, and the weight of Lucas-Kanade＇s method is defined as the reciprocal of the conditioning number of its Hessian matrix. This can eliminate the uncertain constrains in part adjacent domains and improve the numerical stability of the solution of the gradient constraint equation. Then the direction histogram is built based on the concept of gray histogram, and optical flow features are extracted. The direction histogram comes from quantifying optical flow. Then a template base and its index sequences are constructed with the direction histogram. Finally, the similarity of different movement sequences is calculated using the sequence alignment, when the maximal similarity of the online sequence and the reference sequence is bigger than a threshold, the movement recognition module can be considered as the movement of the current sequence is the same as the reference sequence,②In this paper, some experiments on several different simple movements are carded out and also prove the effectiveness of the above method.

关 键 词：光流场 序列比对 动作识别 

分 类 号：R318.6[医药卫生—生物医学工程]