RESUMO
Intensive and lasting stress may induce severe damage to a human's physical and mental health. Successful stress management depends on the effective monitoring of people's everyday activities, in particular, their sedentary behaviors. Here, we propose an unobtrusive office sedentary behavior monitoring system that combines Bluetooth signals and ballistocardiogram (BCG) signals to classify an individual's sitting modes into four categories: off-seat, sedate, working, and in-motion. The proposed monitoring system simultaneously reads received signal strength indicators (RSSI) from several fixed Bluetooth Low Energy (BLE) beacons and BCG data from the piezoelectric sensor placed underneath the chair cushion, with distinct sampling frequencies. The raw signals are first denoised with local subspace projection. Then we extract the local spectral features from the reconstructed signal and the signal differences for a two-stage stacking learning algorithm. The temporally classified results establish a desk-based worker's sedentary profile and make possible the timely intervention of physical inactivity. We tested the prototype system for 15 subjects, and the preliminary results achieved 95% accuracy, demonstrating its potential in a real-world application.