Dongcheng He¹, Janet H. Hsiao¹; ¹Hong Kong University of Science and Technology

Recent research has suggested that the perception of human movement from point-light displays (PLDs) depends on deriving form from motion. We investigated how form information availability and action familiarity modulate attention behavior and performance in biological motion (BM) prediction through eye tracking. In the experiment, stimuli were 28 short PLDs of human movements depicting either familiar daily actions or unfamiliar, animal-imitating movements. Each consisted of 10 frames (IFI=0.1s) of twelve point-lights (size=0.25 deg) marking three joints on each of four limbs. Stimuli varied by viewpoint (front/side), orientation (±20°, ±40°), and the presence or absence of connecting skeleton lines that provided explicit form information. After a fixation display (1s), participants viewed a stimulus and then reported the speed and direction of a target joint by drawing a matching line with the mouse. Participants (N=12, age:18-27 years) showed faster response times (RTs) (p=0.042) and reported higher speeds (p=0.003) for familiar actions, and the familiarity effect in RT interacted with form information availability (p=0.018), with a larger familiarity effect when form information was unavailable. Using Eye Movement analysis with Hidden Markov Models (EMHMM), we found two distinct eye movement patterns through clustering: a centralized pattern focusing on the stimulus center and a distributed pattern attending to both left and right of the stimulus in addition to the center (with a significant difference; p<0.01). Participants showed a more distributed pattern when form information was unavailable and thus had to be inferred from motion alone (p=0.014). Additionally, participants had less consistent eye movement behavior for unfamiliar actions, indicated by higher HMM entropy than for familiar actions (p=0.024). These findings suggest that the effects of form information availability and action familiarity on BM prediction were reflected in different aspects of eye movement behavior, which may jointly influence BM prediction performance.

Acknowledgements: Many thanks to the Research Grant Council of Hong Kong for supporting this work (AoE/E-601/24-N).