Modeling Temporal Interactions with Interval Temporal Bayesian Networks for Complex Activity Recognition

摘要: Complex activities typically consist of multiple primitive events happening in parallel or sequentially over a period time. Understanding such requires recognizing not only each individual event but, more importantly, capturing their spatiotemporal dependencies different time intervals. Most the current graphical model-based approaches have several limitations. First, time--sliced models as hidden Markov (HMMs) and dynamic Bayesian networks are based on points they hence can capture three temporal relations: precedes, follows, equals. Second, HMMs probabilistic finite-state machines that grow exponentially number increases. Third, other syntactic description-based methods, while rich modeling relationships, do expressive power to uncertainties. To address these issues, we introduce interval network (ITBN), novel model combines Network with algebra explicitly Advanced machine learning methods introduced learn ITBN structure parameters. Experimental results show by reasoning dependencies, proposed leads significantly improved performance when complex involving both sequential events.