A Two-Level Approach to Characterizing Human Activities from Wearable Sensor Data

The rapid emergence of new technologies in recent decades has opened up a world of opportunities for a better understanding of human mobility and behavior. It is now possible to recognize human movements, physical activity and the environments in which they take place. And this can be done with high precision, thanks to miniature sensors integrated into our everyday devices. In this paper, we explore different methodologies for recognizing and characterizing physical activities performed by people wearing new smart devices. Whether it’s smartglasses, smartwatches or smartphones, we show that each of these specialized wearables has a role to play in interpreting and monitoring moments in a user's life. In particular, we propose an approach that splits the concept of physical activity into two sub-categories that we call micro- and macro-activities. Micro- and macro-activities are supposed to have functional relationship with each other and should therefore help to better understand activities on a larger scale. Then, for each of these levels, we show different methods of collecting, interpreting and evaluating data from different sensor sources. Based on a sensing system we have developed using smart devices, we build two data sets before analyzing how to recognize such activities. Finally, we show different interactions and combinations between these scales and demonstrate that they have the potential to lead to new classes of applications, involving authentication or user profiling.