Abstract
Canonical-views for static objects are defined as preferred-views that provide richer spatial information and richer cognitive processing (Palmer, Rosch, & Chase, 1981; Blanz, Tarr, & Bülthoff, 1999; Ghose & Liu, 2013). For simple dynamic scenes consisting of three moving balls on a flat surface, canonical-views provide richer information about temporal changes in the spatial relationships (Garsoffky, Huff, & Schwan, 2007; Garsoffky, Schwan, & Huff, 2009). In this research we ask (1) if there is a single canonical-view for an entire movie depicting an assembly task, and (2) if the canonical-view depends on a specific assembly-step or action-type. The stimuli consisted of three complex assembly tasks taken from a technical construction toolkit. Each scenario was recorded from 5 different viewpoints, namely, self, opposite-to-self, left, right and top. The mean assembly time per scenario was 4 min and 14 sec. The actor in the video was right-handed. There were 66 assembly-steps with 7 action-types. Procedure: A pair of video clips showing the same step appeared successively at two different positions on the screen. After the first showing, the videos were played back in a loop until the participant responded. The task was to choose the “better-view” for learning the assembly-step. Data from 20 participants show that canonical-view varies to some extent from assembly-step to assembly-step. The self and the opposite-to-self views were most-preferred 74% of the time. Left and right views were least-preferred 76% of the time. That is, views along the median plane were rated as “better” for learning than frontal and top plane for every action-type. An “unambiguity measure” was calculated for each assembly-step as mean frequency of (most minus least) preferred-views. Higher unambiguity values mean that the best perspective is unambiguously preferred. The median views are unambiguously preferred over other views suggesting these are canonical-views of actions.
Meeting abstract presented at VSS 2015