Distributed activity recognition using key sensors

Recent development of sensor technology gives us the opportunity to effectively monitor daily activities of individuals. As such, in this paper we present a distributed technique to recognize activities of daily living (ADLs) using simple sensors. We consider a number of randomly deployed sensors in home environment augmented with home appliances (e.g., cabinet, desk, chair etc.). Our proposal consists of three major steps. At first, in a random arrangement of sensors, their triggering pattern under human actions is recorded. These records are assembled for meaningful information. This is followed by the categorization of the key sensors (i.e., most important sensors) for each activity from the acquired knowledge. Finally, we group the sensors such that activity based hierarchical clusters can be formed. The system is thus ready for activity recognition. Experiments reveal that even for a small dataset, our proposal can find out the key sensors and form clusters. Also, it is observed that our proposed mechanism yields an accuracy of determination is more than 61%. In addition, it ensures distribution of processing loads among the sensors themselves and thus minimizes the centralized processing overheads.