基于加速度轨迹图像的手势特征提取与识别

针对手势加速度识别中存在数据维度高、计算量大等问题,提出一种基于加速度轨迹图像的手势NMF特征提取与识别方法。通过Wiimote手柄获取手势动作的加速度信号,经过实时有效手势动作分割后,将加速度数据转换为手势轨迹图像,并用非负矩阵分解对手势加速度轨迹图像提取特征向量,最后构建离散隐马尔可夫模型实现目标手势识别。加速度手势轨迹图像转换及采用非负矩阵分解的特征提取方法将未知手势轨迹特征转换为低维子特征序列,降低了计算复杂度,实验表明,该方法能有效识别手势动作。...

Full description

Saved in:
Bibliographic Details
Published in计算机应用研究 Vol. 34; no. 3; pp. 924 - 927
Main Author 刘蓉 刘家祺 刘红
Format Journal Article
LanguageChinese
Published 华中师范大学物理科学与技术学院,武汉,430079%华中师范大学计算机学院,武汉,430079 2017
Subjects
人机交互
加速度传感器
手势识别
隐马尔可夫模型
非负矩阵分解
加速度传感器
human-computer interaction
gesture recognition
accelerometer
人机交互
隐马尔可夫模型
non-negative Matrix factorization (NMF)
discrete hidden Markov model (DHMM)
非负矩阵分解
手势识别
Online AccessGet full text
ISSN1001-3695
DOI10.3969/j.issn.1001-3695.2017.03.066

Cover

More Information
Summary:针对手势加速度识别中存在数据维度高、计算量大等问题,提出一种基于加速度轨迹图像的手势NMF特征提取与识别方法。通过Wiimote手柄获取手势动作的加速度信号,经过实时有效手势动作分割后,将加速度数据转换为手势轨迹图像,并用非负矩阵分解对手势加速度轨迹图像提取特征向量,最后构建离散隐马尔可夫模型实现目标手势识别。加速度手势轨迹图像转换及采用非负矩阵分解的特征提取方法将未知手势轨迹特征转换为低维子特征序列,降低了计算复杂度,实验表明,该方法能有效识别手势动作。
Bibliography:51-1196/TP
gesture recognition; accelerometer; non-negative Matrix factorization（NMF）; discrete hidden Markov model（DHMM）; human-computer interaction
Liu Rong1, Liu Jiaqi2, Liu Hong1（ 1. College of Physical Science ＆ Technology, 2. College of Computer Science, Central China Normal University, Wuhan 430079, China）
There exits high computational complexity and time complexity on accelerometer-based gesture recognition. This paper proposed a gesture NMF features extraction and recognition scheme based on acceleration tracks. First of all,it captured the gesture acceleration data by a Wiimote controller,after real-time gesture segmentation by using the gesture segmentation algorithm based on slope threshold and error threshold of fitting lines,and transformed the gesture acceleration data into gesture track images,it used non-negative matrix factorization for extracting feature vectors,and used discrete hidden Markov model for target recognition. In this proposed method,high-dimensional gesture acceleration data wer
ISSN:1001-3695
DOI:10.3969/j.issn.1001-3695.2017.03.066