Action recognition is viewpoint-dependent in the visual periphery

Abstract

Recognizing actions of others across the whole visual field is required for social interactions. In a previous study, we have shown that recognition is very good even when life-size avatars who were facing the observer carried out actions (e.g. waving) and were presented very far away from the fovea (Fademrecht, Bülthoff, & de la Rosa, 2016). We explored the possibility whether this remarkable performance was owed to life-size avatars facing the observer, which - according to some social cognitive theories (e.g. Schilbach et al., 2013) - could potentially activate different social perceptual processes as profile facing avatars. Participants therefore viewed a life-size stick figure avatar that carried out motion-captured social actions (greeting actions: handshake, hugging, waving; attacking actions: slapping, punching and kicking) in frontal and profile view. Participants' task was to identify the actions as 'greeting' or as 'attack' or to assess the emotional valence of the actions. While recognition accuracy for frontal and profile views did not differ, reaction times were significantly faster in general for profile views (i.e. the moving avatar was seen profile on) than for frontal views (i.e. the action was directed toward the observer). Our results suggest that the remarkable well action recognition performance in the visual periphery was not owed to a more socially engaging front facing view. Although action recognition seems to depend on viewpoint, action recognition in general remains remarkable accurate even far into the visual periphery.

Action Recognition in a Crowded Environment

“…This measure, averaged across all crowd members, represents the amount of motion visible in the image plane calculated across two movie frames. We have shown previously that this measure of motion correlates with the 2D motion energy of the stimuli ( Fademrecht et al., 2017 ). The analysis shows that the amount of movement is indeed significantly higher in the active crowd than in the idle crowd, t (100.59) = 13.91, p < .001, across the movie frames.…”

“…Moreover, action recognition performance in the visual periphery as probed by life-size human stick figures (see example in Figure 3 ) is different from action recognition performance probed by small point-light humans. For example, we have previously shown that action recognition of life-size stick figures (∼32° visual angle [VA]) is excellent up to 75° eccentricity ( Fademrecht, Bülthoff, & de la Rosa, 2016 , 2017 ). In contrast, the detection and discrimination of small point-light walkers (11° VA) was significantly decreased already at 12° eccentricities ( Ikeda, Blake, & Watanabe, 2005 ; Ikeda, Watanabe, & Cavanagh, 2013 ; Thompson, Hansen, Hess, & Troje, 2007 ).…”

“…We used a setup in which actions were carried out by life-size human stick figures to provide both form and motion information of an action. Furthermore, our display allowed the assessment of action discrimination across the entire horizontal visual field (for more details, see Fademrecht et al, 2016Fademrecht et al, , 2017.…”

Human Action Recognition Based on RGB-D and Local Interactive Regions Detection

“…However, by analysis one may find that in some actions such as drinking, the recognition of interactive region between human and the cup can also provide help for action identification. Since the two-dimensional RGB image is the projection of three-dimensional objects [9], interaction detection in the traditional color images cannot effectively reflect the spatial relationship between human and the object, which will result in error extraction [13][14]. At this point, the depth image has a significant advantage [15].…”

Action Recognition using Key-Frame Features of Depth Sequence and ELM

“…different actions captured from one same person. These difficult issues constraint the further progress of vide technology based on RGB sequences [6], [7]. However, the release of Kinect sensor presents a new idea and more information to resolve these problems [8]- [10].…”