Xiaojie Yu / China University of Mining and Technology
Fall detection plays a critical role in home care for the elderly. In this paper, we propose AP-Fall, a novel system that can perform robust fall detection via active and passive acoustic sensing relying solely on a home audio device. Particularly, the speaker is triggered to emit 20kHz wave only when a person walking is detected. We first separate the signal recorded by the microphone into inaudible signal and audible signal. Then we employ the double-checked event detection mechanism to eliminate possible false alarms. A series of fall-related features are extracted and adaptively fused based on the signal-to-noise ratio (SNR) for robust fall detection. We implement AP-Fall and evaluate its performance extensively. Results show that AP-Fall has a better fall detection accuracy of around 94\% in most common circumstances..