Concurrent Recognition of Cross-Scale Activities via Sensorless Sensing

Existing activity recognition focuses on a special scale activity with special sensors. In recent years, the sensorless sensing method which uses wireless signal to recognize activities by refraction and reflection of human body has been widely concerned. However, human activities in real situations are often concurrent (e.g., breathing is associated with motion state). These observations motivate us to answer the question of how to effectively recognize an activity that relies on concurrent cross-scale activities in sensorless pattern. To understand and attack this problem, this paper takes the daily activity recognition of drinking water into account. On the one hand, the measurement of drinking water quantity is of great significance to the determination of hydration state of human body. On the other hand, the micro-scale movements (i.e., respiration and deglutition) and meso-scale actions (i.e., arm movements) during drinking are typical concurrent cross-scale activities. A concurrent cross-scale activities recognition system is designed and implemented by taking the advantage of the ubiquitous channel state information, namely, DW-health, which can detect breath and drinking, estimate the water intake with commercial Wi-Fi devices. DW-health adopts the novel filter design which engenders better performance of noise elimination. Moreover, the activity segmentation method based on subcarrier correlation and dynamic threshold is proposed to recognize concurrent recognition of cross-scale activities. The experimental results indicate that the drinking and non-drinking detection accuracy of DW-health can reach 97% and 84.83%, respectively.