Composite event recognition (CER) frameworks reason over streams of low-level, symbolic events in order to detect instances of spatio-temporal patterns defining high-level, composite activities. The Event Calculus is a temporal, logical formalism that has been used to define composite activities in CER, while RTEC∘ is a formal CER framework that detects composite activities based on their Event Calculus definitions. RTEC∘, however, cannot handle arbitrary Event Calculus definitions for composite activities, limiting the range of CER applications supported by RTEC∘. We propose RTEC, an extension of RTEC∘ that supports arbitrary composite activity specifications in the Event Calculus. We present the syntax, semantics, reasoning algorithms and time complexity of RTEC. Moreover, we propose a compiler for RTEC, generating the optimal representation of an input set of Event Calculus definitions. We demonstrate the correctness of our compiler and outline its time complexity. We conducted an empirical evaluation of RTEC on synthetic and real data streams from human activity recognition and maritime situational awareness, including a comparison with two state-of-the-art Event Calculus-based systems, which demonstrates the benefits of RTEC.
