基于函数型数据分析方法的人体动态行为识别

发布时间：2018-02-24 01:09

本文关键词： 动态行为识别连续性与周期性周期行为函数型数据分析可穿戴式运动捕捉系统　出处：《自动化学报》2017年05期 　论文类型：期刊论文

【摘要】：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%.
[Abstract]:In order to better analyze the continuity and periodicity of human daily motion behavior, the limb motion is continuous and the corresponding motion capture data are discrete. In this paper, a human body dynamic behavior recognition method based on functional data analysis is proposed. Firstly, the human periodic behavior data collected by wearable motion capture system is functioned by using functional data analysis method. The continuity and periodicity of the data are accurately defined by the function. Then, the starting point of a motion period is determined according to the derivative function information, and the data sequence of a motion period is approximately extracted. Finally, According to the difference of curve characteristics in a period of different behaviors, the support vector machine is used to classify and recognize the dynamic behavior. The experimental results show that the proposed algorithm can describe the continuity and periodicity of human dynamic behavior. The motion data is aligned at the unified starting point of calibration, and the recognition rate of the WARD data set and the self-collected data set are better, reaching 97.5% and 98.75 respectively.
【作者单位】：安庆师范大学计算机与信息学院;智能感知与计算安徽省高校重点实验室;安庆师范大学数学与计算科学学院;
【基金】：国家自然科学基金(11471093) 国家科技支撑课题(2014BAH13F02) 安徽省高校自然科学研究重点项目(KJ2014A142) 情感计算与先进智能机器安徽省重点实验室开放课题(ACAIM160102)资助~~
【分类号】：O213;TP18;TP212.9

【参考文献】