基于函数型数据分析方法的人体动态行为识别

Human Dynamic Action Recognition Based on Functional Data Analysis

人体运动过程中,肢体的运动是连续的,而对应的运动捕捉数据是离散的.为了更好地分析人体日常运动行为的连续性与周期性,本文提出了一种基于函数型数据分析(Functional data analysis,FDA)的人体动态行为识别方法.首先,利用函数型数据分析方法,将可穿戴式运动捕捉系统采集的人体周期行为数据函数化,通过函数准确地定义数据的连续性与周期性;然后,根据导函数信息确定一个运动周期的起始点,并近似地提取出一个运动周期的数据序列;最后,根据不同行为一个周期内的曲线特征差异,利用支持向量机对动态行为进行分类识别.实验结果表明,本文的算法既能够较好地描述人体动态行为的连续性与周期性,又使得运动数据在标定的统一起始点处对齐,且在WARD数据集与自采集数据集上均取得了较好的识别率,分别达到97.5%与98.75%.

In human motion, limb movement is continuous. However, the corresponding motion capture data is discrete. This paper explores a method for human dynamic action recognition based on functional data analysis （FDA） so as to analyze the continuity and periodicity of daily action. Firstly, we transform the periodic data collected by the wearable motion capture system into functional data using FDA, and then define the continuity and periodicity of data exactly by using function properties. Secondly, we determine the initial point of a motion period according to the derivative information, and then extract the data series representing a period of motion. Finally, we utilize support vector machine （SVM） to classify the dynamic action according to the different characteristics of the curves about different actions in a period. The experimental result indicates that our algorithm can describe the continuity and periodicity of hmnan dynamic action, and align the motion data at the uniform start point we determined. At the same time, desirable recognition rates, such as 97.5 % and 98.75 %, can be achieved based on WARD and our database using our algorithm.