[1]何继爱,李先祺,赵雪.一种室内WiFi环境下的人体活动识别方法[J].计算机技术与发展,2025,(03):179-186.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0345]
　HE Ji-ai,LI Xian-qi,ZHAO Xue.A Human Activity Recognition Method in Indoor WiFi Environment[J].,2025,(03):179-186.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0345]
点击复制

一种室内WiFi环境下的人体活动识别方法()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年03期
页码:
179-186
栏目:
新型计算应用系统
出版日期:
2025-03-10

文章信息/Info

Title:
A Human Activity Recognition Method in Indoor WiFi Environment
文章编号:
1673-629X(2025)03-0179-08
作者:
何继爱; 李先祺; 赵雪
兰州理工大学 计算机与通信学院,甘肃 兰州 730050
Author(s):
HE Ji-ai; LI Xian-qi; ZHAO Xue
School of Computer and Communication,Lanzhou University of Technology,Lanzhou 730050,China
关键词:
无线感知; 人体活动识别; 信道状态信息; 机器学习; 优化算法
Keywords:
wireless perception; human activity recognition; channel status information; machine learning; optimization algorithm
分类号:
TP391
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0345
摘要:
针对人体活动识别邻域中的隐私保护性差、识别准确率低、成本高等问题,提出一种基于 WiFi 信号识别人体活动的方法,该方法利用人体活动对 WiFi 信道状态信息的影响实现非接触式的活动识别。 首先,利用子载波的相关性及其对活动敏感性的差异提出一种天线-子载波选择策略,减少不敏感天线及子载波对活动识别的影响;然后,对包含活动特征的 WiFi 信道状态信息用 Hampel 滤波及小波变换去噪进行预处理,并利用滑动窗口内信号的方差变化确定活动区间,去除不含活动特征的冗余信息,再用统计特征构造用于分类的特征量;最后,利用麻雀搜索算法优化正则化学习机的参数选择过程以提高模型性能。 实验结果表明,该方法对拍手、前踢、深蹲、步行、弯腰、打电话、坐下、高摆手、喝水九种人体动作平均准确率可达到96% ,所提出的天线-子载波选择策略将九种活动的准确率平均提高了4. 56% ;通过与目前先进算法和其他改进算法的对比,有效证明了该方法的有效性。
Abstract:
Aiming at the problems of poor privacy protection,low recognition accuracy and high cost in the human activity recognition neighbourhood,a method for human activity recognition based on WiFi signals is proposed,which uses the effect of human activity on WiFi channel state information to achieve non-contact activity recognition. Firstly,an antenna-subcarrier selection strategy is proposed using the correlation of subcarriers and their differences in activity sensitivity to reduce the impact of insensitive antennas and subcarriers on activity recognition. Then,WiFi channel state information containing activity features is pre - processed with Hampel filtering and wavelet transform denoising,and the variance change of signals within the sliding window is used to determine the activity interval,and the redundant information without activity features is removed,and statistical features are used to construct the feature volume for classifi-cation. Finally,the regularized learning machine parameter selection process is optimized using a sparrow search algorithm to improve the model performance. The experimental results show that the proposed method can reach an average accuracy of 96% for nine human body movements:clapping, front kicking, deep squatting, walking, bending, talking on the phone, sitting down, high hand swinging, and drinking water,and the proposed antenna - subcarrier selection strategy improves the accuracy of the nine activities by an average of 4.56% . By comparing with the current state-of-the-art algorithms and other improved algorithms,the validity of the proposed method has been effectively proved.
更新日期/Last Update: 2025-03-10