||Distributed ambient and on-body sensor systems can provide a suitable basis for recognizing complex human activities in daily life. Moreover, distributed activity recognition systems have high prospects for handling processing and communication loads more effectively than centralized solutions. A key challenge is to construct distributed activity recognition systems that make efficient use of the resources available for the recognition task, considering scalability and dynamic system reconfiguration. In this work, we present an approach to distributed activity recognition by introducing an activity-event-detector (AED) concept. We show formally how to construct and use AED for distributed recognition systems based on directed acyclic graphs. We illustrate essential properties for system scalability and efficiency using AED graphs. Results from a home monitoring study targeted at monitoring daily life activities are presented to illustrate the AED-based model regarding applicability and reconfiguration.