摘要
针对现有依靠穿戴设备、雷达和视频图像人体行为感知的方法对环境要求高,成本高,且不利于保护隐私等问题,提出一种基于信道状态信息无设备且低成本的日常行为识别方法。通过商用WiFi设备采集原始CSI数据,在无需信号进行去噪处理的情况下通过提取原始CSI最大程度能提高识别精度的三阶累积量特征,应用基于互信息的特征选择算法(MIFS)对特征进行筛选,将筛选得到的特征子集输入进粒子群优化算法(PSO)优化过参数的支持向量机(SVM)分类器中以测量性能。实验结果表明,该方法在不同环境下对不同日常动作的平均识别率达到了96.1%,验证了该方法用于行为识别能获得较高的准确率和鲁棒性。
The existing methods rely on wearable devices,radar and video image human behavior perception,high environmental requirements,the cost is high and it is not conducive to protecting privacy and other issues.Raw CSI data were collected through commercial WiFi equipment,the third-order cumulant feature that could maximize the recognition accuracy was extracted by extracting the original CSI without signal denoising processing,and the feature selection algorithm based on mutual information(MIFS)was applied to the feature perform screening,and the selected feature subsets were inputted into the particle swarm optimization algorithm(PSO)optimized support vector machine(SVM)classifier to measure performance.Experimental results show that the average recognition rate of proposed method for different daily actions in different environments reaches 96.1%,it is verified that the method can obtain high accuracy and robustness for behavior recognition.
作者
周祥
常俊
武浩
ZHOU Xiang;CHANG Jun;WU Hao(School of Information Science and Engineering,Yunnan University,Kunming 650091,China)
出处
《计算机工程与设计》
北大核心
2022年第1期231-236,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61162406)
云南省省教育厅科研基金项目(2019J0007)。
关键词
信道状态信息
行为识别
三阶累积量
互信息
特征选择
粒子群优化
支持向量机
channel state information
behavior perception
third-order cumulant
mutual information
feature selection
particle swarm optimization
support vector machine
