摘要： 传感器与摄像头等设备的传统动作识别存在受环境影响大及侵犯用户隐私等问题，以京剧动作为研究对象，提出一种非接触式人员动作识别方法Wi-Opera。在离线阶段采集Wi-Fi路由设备上人体动作的信道状态信息（CSI）数据，利用巴特沃斯低通滤波器和小波变换方法对CSI数据分别进行去噪和平滑处理，通过主成分分析算法提取动作的特征值构建每个京剧动作的决策树，最终形成随机森林模型。在在线阶段实时采集的动作数据经过处理后，将京剧动作的特征值输入随机森林模型中进行识别，从而输出识别结果。实验结果表明，Wi-Opera方法的综合识别精度为94.6%，具有较高的识别精度和较强的鲁棒性。

关键词: 信道状态信息, 人员动作识别, 巴特沃斯低通滤波器, 小波变换, 主成分分析, 随机森林模型

Abstract: The performance of traditional sensors and cameras in motion recognition are affected by the environmental factors and limited by the risk of violating user privacy.Taking the Peking Opera actions as the research object, this paper proposes a non-contact human motion recognition method, Wi-Opera.In the offline phase, the data of Channel State Information(CSI) of human actions is collected from Wi-Fi routing devices, and then denoised as well as smoothed by using the Butterworth low-pass filter and the wavelet transform method.On this basis, the Principal Component Analysis(PCA) algorithm is used to extract the eigenvalues of the actions to construct a decision tree for each Peking Opera action, and finally form a Random Forest Model(RFM).In the online phase, after the action data collected in real time is processed, the eigenvalues of Peking Opera actions are input into the RFM for recognition, and the recognition result is output.Experimental results show that the comprehensive recognition accuracy of the Wi-Opera method is 94.6%, and the method has high recognition accuracy and strong robustness.

Key words: Channel State Information(CSI), human motion recognition, Butterworth low-pass filter, wavelet transform, Principal Component Analysis(PCA), Random Forest Model(RFM)

中图分类号: 

• TP391