Adaptive energy-saving strategy for activity recognition on mobile phone

Most existing mobile devices nowadays are powered by a limited energy resource. With the tendency using machine learning on mobile devices for activity recognition (AR), recent achievements still remain restrictions including low accuracy and lacking of evidences about power consumption of feature extraction and classification. Moreover, keeping constantly a high sampling frequency was the most power consuming factor. In this paper, we contribute a novel method for extracting features in time domain and frequency domain. These features are then classified by Support Vector Machine (SVM). Prototypes of the proposed methods are then implemented on a cell phone to measure power consumptions. To reduce the energy overhead of continuous activity recognizing, we propose an adaptive energy-saving strategy by selecting an appropriate combination of flexible frequency and classification feature for each individual. The self-construct data and SCUTT-NAA dataset are used in our experiment. We achieved an overall 28 percent of energy saving in activity recognition on mobile phone.