Human activity recognition is essential for various smart-home applications. With the development of sensing technology, various approaches have been proposed for occupancy monitoring indoors. However, such approaches have practical limitations that they require additional occupancy sensors, which may raise privacy issues and obtrude on occupants’ daily lives. In this research, a Wi-Fi-based occupancy monitoring system, Wi-Sensing, is proposed to recognize occupant’s activities of daily living in a non-intrusive way by exploiting commercial off-the-shelf Wi-Fi devices. Channel State Information (CSI) has been extracted from Wi-Fi signals collected from multiple Wi-Fi devices, which could be replaced by Internet of Things (IoT) devices. While multiple receivers are needed to cover the entirety of an indoor space, previous approaches have been proposed to extract numerous features from a single transmitter–receiver pair. In this context, this study presents a new approach toward extracting spatial–temporal features from multiple receivers deployed throughout an indoor space. In this approach, a Short-Time Fourier Transform (STFT) was used to convert time-series CSI data into image data. The converted image data from each receiver was then integrated as large image data, which preserved the temporal-spatial information of all the receiver data. A Convolutional Neural Network (CNN) was used as a feature extractor for the image data, and Long Short-Term Memory (LSTM) was exploited to classify basic activities in daily life (e.g., personal hygiene, eating, mobility, etc.). Wi-Sensing provides over 96% classification accuracy in two different indoor environments.

人类活动识别对于各种智能家居应用至关重要。随着传感技术的发展，已经提出了各种用于室内占用监测的方法。然而，这种方法具有实际的局限性，即它们需要附加的占用传感器，这可能会引起隐私问题并妨碍占用者的日常生活。在这项研究中，提出了一种基于Wi-Fi的占用监测系统Wi-Sensing，该漏洞可通过利用现成的商用Wi-Fi设备以非侵入性的方式来识别占用者的日常生活。通道状态信息（CSI）是从多个Wi-Fi设备收集的Wi-Fi信号中提取的，可以由物联网（IoT）设备代替。虽然需要多个接收器来覆盖整个室内空间，已经提出了先前的方法来从单个发送器-接收器对中提取众多功能。在这种情况下，本研究提出了一种新方法，可从部署在整个室内空间的多个接收器中提取时空特征。在这种方法中，使用了短时傅立叶变换（STFT）将时序CSI数据转换为图像数据。然后将来自每个接收器的转换后的图像数据整合为大图像数据，从而保留所有接收器数据的时空信息。卷积神经网络（CNN）被用作图像数据的特征提取器，而长期短期记忆（LSTM）被用来对日常生活中的基本活动进行分类（例如，个人卫生，饮食，活动能力等）。