Commercial Wi-Fi Based Fall Detection with Environment Influence Mitigation

Abstract

Motivated by the urgent demands of indoor fall detection, significant progress has been made to explore Wi-Fi based fall detection techniques that utilize the information collected by commercial Wi-Fi signals to infer falling without carrying a dedicated device. Existing commercial Wi-Fi based fall detection systems, though yielding reasonably good performance in certain cases, are faced with a major challenge. The Wi-Fi signals arriving at the receiving devices usually carry substantial information specific to the environment where falling happens. Due to this reason, a fall detection model trained in one specific environment does not work well in the other. Furthermore, it is labor-intensive and time-consuming to acquire sufficient data and rebuild fall detection model in each new environment. To address this challenge, we propose TL-Fall, a transfer-learning based device free fall detection system. Specifically, the fall detection model is trained with labeled data to derive knowledge in the old environment. With the derived knowledge, the fall detection model working in the new environment can be trained with only a few labeled data. The effective across environment knowledge reuse mitigates the environmental influence and maintains the satisfied fall detection accuracy as well. The extensive experiments are conducted on TL-Fall in typical indoor environment. The experimental results demonstrate the superior effectiveness of TL-Fall, with 86.83% fall detection sensitivity and 84.71% fall detection specificity on average.