The next generation of health activity monitoring is greatly dependent on wireless sensing. By analysing variations in channel state information, several studies were capable of detecting activities in an indoor setting. This paper presents promising results of an experiment conducted to identify the activity performed by a subject and where it took place within the activity region. The system utilises two Universal Software Radio Peripheral (USRP) devices, operating as software-defined radios, to collect a total of 360 data samples that represent five different activities and an empty room. The five activities were performed in three different zones, resulting in 15 classes and a 16t h class representing the room whilst it is empty. Using the Random Forest classifier, the system was capable of differentiating between the majority of activities, across the 16 classes, with an accuracy of almost 94 %. Moreover, it was capable of detecting whether the room is occupied, with an accuracy of 100 %, and identify the walking directions of a human subject in three different positions within the room, with an accuracy of 90 %
The next generation of health activity monitoring is greatly dependent on wireless sensing. By analysing variations in channel state information, several studies were capable of detecting activities in an indoor setting. This paper presents promising results of an experiment conducted to identify the activity performed by a subject and where it took place within the activity region. The system utilises two Universal Software Radio Peripheral (USRP) devices, operating as software-defined radios, to collect a total of 360 data samples that represent five different activities and an empty room. The five activities were performed in three different zones, resulting in 15 classes and a 16t h class representing the room whilst it is empty. Using the Random Forest classifier, the system was capable of differentiating between the majority of activities, across the 16 classes, with an accuracy of almost 94 %. Moreover, it was capable of detecting whether the room is occupied, with an accuracy of 100 %, and identify the walking directions of a human subject in three different positions within the room, with an accuracy of 90 %