Skip to Main Content
In this paper we demonstrate the application of a probabilistic grammar-based formulation to detect complex activities from simple sensor measurements. In particular, we present a grammar hierarchy for identifying "cooking activity" from low-level location measurements in an assisted living application. Using real data from a pilot network deployment, we show that our system can recognize complex behaviors in a manner that is invariant across multiple different instances of the same activity. Our experiments also demonstrate that substantial data interpretation can take place at the node level, allowing the network to operate on compact symbolic representations.