Mapping of Activity Recognition as a Distributed Inference Problem in Sensor Network.

Abstract- This paper presents a distributed model for detecting Activities of Daily Living (ADLs) in a home setting. We consider an environment where household devices and utensils are augmented with simple sensors. The variation of object usage and shortage of training data makes it infeasible to construct high level models for ADLs in this environment. However, ADLs produce a series of sensor activations and the sensor activation sequences can help identifying the activity being performed. Hence, we try to correlate the sensor activation sequences with the activities. We provide a simple distributed linear time inference algorithm for this. The algorithm scales well with the size of the sensor network and number of activities to be detected.