Towards a cognitive-inspired automatic unitizing technique: a feasibility study

Abstract

In this paper, we present and assess a novel technique for unitizing inspired by a cognitive theory on event structure perception. Unitizing indicates the process of dividing an observation into smaller units. Unitizing is often performed automatically, e.g., by selecting fixed-length windows. Although fast, such approach might result in unit boundaries being placed mid-interaction, eventually affecting observation, annotation, and labeling. We conceived a unitizing technique based on the Event Segmentation theory. In brief, changes drive the perception of boundaries between events (or units): an unexpected change in the observed situation might mean the current event ended and a new one begun. Our technique relies on observed changes for identifying unit boundaries. The first sketch of our technique was recently tested, proving it effective in overcoming the aforementioned shortcomings of fixed-window unitizing. Here, we further explore its feasibility by testing it in a different domain, i.e., solo stage performances, in order to explore the feasibility of adopting our unitizing approach across domains. Our results further support the idea of leveraging the Event Segmentation Theory for the design of an automatic technique for video unitizing.