부산대학교 도서관

The combination of smart sensors and affective computing capabilities in wearable devices enables future technological integration horizons for high added value applications. Among the usual information considered in the field of affective computing, those based on physiology have gained special attention in recent years, since it is related to the autonomic nervous system (ANS), which is responsible for physiological regulation for stress and relaxed situations. One usual physiological metric is heart rate variability (HRV), from which information related to ANS activation can be extracted. The analog front end circuitry for physiological smart sensors is facing a revolution including not only signal conditioning but signal processing capabilities as well. However, despite the efficiency offered by the sensors, an exhaustive design space exploration (DSE) for every sensor involved within the system is recommended to maximize the embedded resource usage. This paper presents a detailed DSE for every stage involved in an HRV based wearable affective computing device developed by the authors. Different signal processing elements are implemented, resulting in a collection of recommendations based on particular wearable applications needs, such as inference time and accuracy of useful affective information extracted. A particular continuous rapid inference use case by considering the DSE recommendations is implemented. This application reaches adequate precision for detecting stress by using only four second temporal processing window.