Passive Human Sensing with COTS Wi-Fi

Recent advances in wireless sensing show promise for ubiquitous human activity recognition interface with Wi-Fi. Despite its attractiveness, multiple challenges still exist in bottom-up translation from CSI values to human activities, making existing approaches unready for practical use. To this end, we recognize three key challenges, non-coherent CSI measurements, context-dependent features and diverse human activities. Our preliminary results demonstrate the feasibility of CSI cleaning and passive human tracking. Based on the initial efforts, we propose the plan for further research.