In the era of the Internet-of-Things (IoT), billions of smart devices are deployed in indoor environments, connect, share data, and integrate information to fulfill users' needs. Wi-Fi is the ubiquitous communication interface in IoT networks. Inspired by the fact that the Wi-Fi signal can interact with the environment during the propagation, it can extend its role from a communication medium to a wireless sensing tool to perceive human activities in surrounding environments. By analyzing the dynamic components of the Wi-Fi radios introduced by human motion, many applications on activity monitoring and detection are enabled. To contribute to the novel applications of Wi-Fi, this dissertation mainly focuses on passive fall detection, indoor proximity detection, and virtual keyboard implementation for Wi-Fi sensing.

In the first part of this dissertation, we propose a novel Wi-Fi-based environment-independent indoor fall detection system by leveraging the features inherently associated with human falls — the patterns of speed and acceleration over time. The system consists of an offline template-generating stage and an online decision-making stage. In the offline stage, the speed of human falls is first estimated based on the statistical modeling about the channel state information (CSI). Dynamic time warping (DTW) based algorithms are applied to generate a representative template for typical human falls. Then fall event is detected in the online stage by evaluating the similarity between the patterns of real-time speed/acceleration estimates and the representative template. Results of extensive experiments demonstrate the proposed system can achieve consistently high accuracy in time-varying line-of-sight (LOS) and non-line-of-sight (NLOS) environments and can be generalized to new environments without re-training.

In the second and third parts, we investigate the feasibility of detecting motion in proximity robustly and responsively based on a single pair of commercial Wi-Fi devices. We establish the connection between the underlying radio propagation properties and the proposed features. Extensive experiments in various environments validate the efficiency of the devised feature-based detection scheme. Further, we generalize the system to a multi-device structure and conduct experiments under single-user and multi-user sensing scenarios. The results verify that responsive on-device proximity detection can be achieved by combining the information from different links, illustrating its potential for real-time home automation applications.

The last part of the dissertation considers the design of a universal virtual keyboard that reuses a commodity 60GHz Wi-Fi radio as a radar. By leveraging the unique advantages of 60GHz Wi-Fi signals, the proposed system can convert any flat surface into an effective typing media and support customized keyboard layouts. We devise a novel signal processing pipeline to detect, segment and separate, and finally recognize keystrokes. The proposed virtual keyboard system enables concurrent keystrokes and does not need any training except for a minimal one-time effort of only three keypresses for keyboard calibration upon the initial setup. Extensive experiments demonstrate a high recognition accuracy for both single-key and multi-key scenarios on different keyboards, presenting the proposed systems as a promising solution to future applications.