Real-Time Human Gaze and Head Pose Detection and Recognition Using Mediapipe

Head posture detection offers a behavioral measure for examining cognition for film viewing. This study investigates how head pose and inferred gaze behavior vary across film genres and how these variations relate to comprehension, satisfaction, mind wandering, and emotional engagement. Participants were instructed to watch curated film clips while their facial and head movements were continuously recorded using a head-pose tracking application. The clips represented five genres: action, comedy, drama, horror, and romance. Collected data were categorized by genre and head orientation, and a 3×3 spatial grid was employed to estimate on-screen gaze regions based on head posture. Heatmaps were generated to visualize attention distribution across scenes. Analysis revealed that, regardless of genre, viewers exhibited diverse gaze patterns; however, specific scenarios, particularly horror sequences, elicited more convergent gaze behavior, suggesting heightened attentional focus. Statistical analysis using the Wilcoxon test demonstrated significant interparticipant differences in head posture patterns. Emotional responses were found to influence gaze direction, with trends closely aligned to the narrative and affective demands of each genre. These findings suggest that head posture–based gaze estimation can provide meaningful insights into audience behavior and emotional engagement, offering a complementary tool for empirical inquiry in film studies.