Evolving activity recognition from sensor streams

Recognizing people´s activity automatically is an important task that needs to be tackled in order to face other more complex tasks such as action prediction, remote health monitoring, or interventions. Recent research on activity recognition has demonstrated that many different activities can be recognized. In most of these researches, the activities are previously predefined as statistic models over time. However, how people perform a specific activity is changing continuously. In this paper we present an approach for classifying different activities from sensor readings based on Evolving Fuzzy Systems (EFS). Thus, the model that describes an activity evolves according to the changes observed in how that activity is performed. This approach has been successfully tested on a real world domain using binary sensors data streams.