Human activity classification using vibration and PIR sensors

Abstract

Fall detection is an important problem for elderly people living independently and people in need of care. In this paper, a fall detection method using seismic and passive infrared (PIR) sensors is proposed. Fast Fourier transform, mel-frequency cepstrum coefficients, and discrete wavelet transform based features are extracted for classification. Seismic signals are classified into “fall” and “not a fall” classes using support vector machines. Once a moving person is detected by the PIR sensor within a region of interest, fall is detected by fusing seismic and PIR sensor decisions. The proposed system is implemented on a standard personal computer and works in real-time.