A feature combination approach for the detection of early morning bathroom activities with wireless sensors

In this paper we investigate the use of wearable accelerometers and wireless home sensors for the detection of early morning daily activities to assist people with cognitive impairments. In particular we focus on the detection of brushing, washing face and shaving activities by using a wireless accelerometer sensor attached to the right wrist of the subjects to collect the hand movement data. We extracted time and frequency domain features of the accelerometer data for activity recognition. In order to compare the efficiency of different frequency domain features, we used fast Fourier transform and autoregressive modeling. The extracted time and frequency domain features are input to an ensemble of Gaussian mixture models (GMM) which represent individual activities we focus on. Finally, they are post processed by a finite state machine for classification. We show promising experimental results from 7 subjects while completing washing face, shaving and brushing activities. The proposed system achieved 93.5%, 92.5% and 95.6% classification accuracy in the recognition of these three tasks respectively.