A Swarm of Wearable Sensors at the Edge of the Cloud for Robust Activity Recognition

Wearable computers intelligently combine data from motion sensors placed at various locations on body with the aim to recognize human activities for the applications of healthcare and wellness. Many activity recognition algorithms for wearable computers exist today. To ensure the effectiveness of the recognition algorithms, the sensors typically have to be worn with a known orientation, since patterns of interest or templates for signal processing would be generated for that orientation. If worn in a disparate orientation, activity recognition algorithms will likely fail. We propose a technique that enables the activity recognition algorithm to function properly irrespective of the orientation of the nodes. This will provide a unique opportunity to assure the effectiveness of the recognition algorithms even when the sensors accidentally move or are misplaced. More importantly, this will enable the notion of reusing data generated in the past potentially by other users, and when the sensors are worn differently. This will eliminate the need for training the system every time it is deployed on a new user for the first time. This feature will be extremely attractive for the swarm of wearable computers capable of generating vast amounts of data. The notion of data reuse will be empowered by performing the proposed technique in the cloud infrastructure or on the wearable computers in real-time.