基于近似熵子载波选择的人体手势识别方法  

Human gesture recognition method based on approximate entropy subcarrier selection

作　　者：田勇[1] 郭莹 崔家栋 李思柔 陈晨 丁学君[2] TIAN Yong;GUO Ying;CUI Jia-dong;LI Si-rou;CHEN Chen;DING Xue-jun(School of Physics and Electronic Technology,Liaoning Normal University,Dalian 116029,China;School of Management Science and Engineering,Dongbei University of Finance and Economics,Dalian 116025,China)

机构地区：[1]辽宁师范大学物理与电子技术学院,辽宁大连116029 [2]东北财经大学管理科学与工程学院,辽宁大连116025

出　　处：《计算机工程与设计》2022年第2期323-329,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(62076114、71874025、71503033);教育部人文社会科学研究基金项目(20YJA630058)。

摘　　要：针对现有基于商用WiFi设备的人体手势识别方法存在的子载波选择不够优化、动作区间截取不够精确等问题,提出一种基于近似熵子载波选择的人体手势识别(AEGR)方法。利用提出的最小近似熵法构建识别方法待处理的CSI幅值数据,对构建的数据采用小波去噪和中值滤波组合法进行去噪;利用滑动窗极差法精确截取CSI幅值的动作区间,据此提取用于分类的8个特征量;利用随机森林算法进行人体手势识别。实验结果表明,AEGR方法的手势识别准确率可达98.75%,验证了其良好性能。Aiming at the problems of suboptimal subcarrier selection and inaccurate action interval interception in existing human gesture recognition methods based on commercial WiFi devices,a human gesture recognition method based on approximate entropy subcarrier selection(AEGR)was proposed.The proposed minimum approximate entropy method was used to construct the CSI amplitude data to be processed using the recognition method,and the constructed data were denoised by the combination of wavelet denoising and median filtering.The motion interval of CSI amplitude was accurately intercepted using the sliding window range method,and eight feature quantities for classification were extracted accordingly.Random forest(RF)classification algorithm was used to recognize human gestures.Results of a large number of experiments show that the recognition accuracy of the proposed AEGR method can reach 98.75%,which verifies its good performance.

关 键 词：手势识别 信道状态信息 近似熵 滑动窗极差 特征提取 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]