Recent advances in the development of wearable sensors and smartphones open up opportunities for executing computing operations on the devices instead of using them for streaming raw data. By minimizing power consumption due to the wireless transmission, limited energy resources of wearable devices can be utilized not only for sensing, but also for processing physiological signals. Computational tasks between a wearable sensor and a smartphone can be distributed efficiently in order to provide balance between power consumption of both processing and transmission of the data. In this paper, we have analyzed the computational balancing between a wearable sensor and a smartphone. Presented models show different trade-offs between classification accuracy, processing time and power consumption due to different number and types of extracted features and classification models. Our results are based on a physiological dataset, where electrocardiogram and electro dermal activity signals were collected from 24 individuals in short-term stress and mental workload detection scenario. Our findings show that placing a feature extraction on a wearable sensor is efficient when processing cost of the extracted features is small. On the other hand, moving classification task to the smartphone can improve accuracy of recognition without compromising the overall power consumption.
展开▼
