Pervasive self-powered human activity recognition without the accelerometer

摘要: Conventional human activity recognition (HAR) relies on accelerometers to frequently sample motion (acceleration). Unfortunately, power consumption of becomes a bottleneck for realising pervasive self-powering HAR as the amount that can be practically harvested from environment is very small. Instead using accelerometer, this paper advocates use energy harvesting signal source when (kinetic) being device. The proposed classifying activities motivated by fact different produce kinetic in way leaving their signatures signal. Using information theoretic analysis experimental data, we show many standard statistical features provide significant gain used discriminating between activities, confirming its potential HAR. We have evaluated accuracy based 14 sets common each containing 2–10 classified. accuracies varied 68% 100% depending set activities. average over all 83%, which within 13% what could achieved with an accelerometer without any constraints.