||Activity-Event-Detector (AED) digraphs can describe relations between human activities, activity-representing pattern events from sensors, and distributed detector nodes. AED graphs have been successfully used to perform network adaptations, including reconfiguring networks to reduce recognition complexity and network energy needs. In this paper, we present an approach to infer AED graph configurations from distributed sensor data. We utilise a non-parametric clustering procedure and derive all relevant information about the AED graph structure, including the detector-specific activity grouping and activity-detector relations from measured data. We analysed our approach using a previously published dataset and compared our inferred AED graph with those designed by an expert. The system based on the inferred AED graph yielded a performance boost of 15% in the final classification accuracy and reduced computational complexity of detectors. These results indicate that our approach is viable to automate the configuration of distributed activity recognition sensor-detector networks.