Activity- and context-aware systems, as they are known, established, and well evaluated in small-scale laboratory settings for years and decades, suffer from the fact, that they are limited concerning the underlying data delivering entities. The sensor systems are usually attached on the body, on objects, or in the environment, directly surrounding persons or groups whose activities or contextual information has to be detected. For sensors that are exploited in this kind of systems, it is essential that their modalities, positions and technical details are initially defined to ensure a stable and accurate system execution. In contrast to that, opportunistic sensing allows for selecting and utilizing sensors, as they happen to be accessible according to their spontaneous availability, without presumably defining the input modalities, on a goal-oriented principle. One major benefit thereby is the capability of utilizing sensors of different kinds and modalities, even immaterial sources of information like webservices, by abstracting low-level access details. This emerges the need to roll out the data federating entity as decentralized collecting point. Cloud-based technologies enable space- and time-free utilization of a vast amount of heterogeneous sensor devices reaching from simple physical devices (e.g., GPS, accelerometers, as they are conventionally included on today's smart phones) to social media sensors, like Facebook, Twitter, or LinkedIn. This paper presents an opportunistic, cloud-based approach for large-scale activity- and context-recognition.