Augmenting Bag-of-Words: Data-Driven Discovery of Temporal and Structural Information for Activity Recognition

摘要

We present data-driven techniques to augment Bag of Words (BoW) models, whichallow for more robust modeling and recognition of complex long-term activities,especially when the structure and topology of the activities are not known apriori. Our approach specifically addresses the limitations of standard BoWapproaches, which fail to represent the underlying temporal and causalinformation that is inherent in activity streams. In addition, we also proposethe use of randomly sampled regular expressions to discover and encode patternsin activities. We demonstrate the effectiveness of our approach in experimentalevaluations where we successfully recognize activities and detect anomalies infour complex datasets.