Using Wi-Fi Signals to Characterize Human Gait for Identification and Activity Monitoring

摘要

Gait characterization and monitoring technologies are useful for the purposes of biometrics tracking and monitoring subjects (e.g., the elderly, at risk, and patients). Traditional techniques of measuring gait employ image processing or special sensors, which require either direct line of sight or physically attached sensors and thus, are cumbersome and costly. We propose Hoble that uses Wi-Fi signals to characterize multipath and Doppler Effect. Because of the physical property, the ubiquity, and the robustness of Wi-Fi signals, this type of sensing penetrates walls and does not require special signals or attachment of sensors to humans. In contrast to previous techniques Hoble 1) extracts features that identify individuals by their intrinsic body movement during walking without attachments to the body, 2) addresses the need to conduct real-time monitoring of individuals and detecting events such as falling, and 3) creates signatures by measuring Channel State Information (CSI), which provide high-fidelity location, movement, and identity information of human subjects. We implemented Hoble on a National Instruments (NI) Radio Frequency (RF) test-bed and conducted extensive experiments on six individuals at three locations. Our empirical results show that by applying a simple Naive Bayes classifier on the extracted features, the correct identification rate was 87%. The correct localization rate was 90%. We demonstrated line-of-sight (LoS) and with non-line-of-sight (NLoS) scenarios.