Projects
Attention on Films

Attention on Films

Overview

How do visuospatial features guide collective attention in films?
📅 Jan 2020 – Dec 2021
🏛️ University of Skövde, Sweden; Örebro University, Sweden; German Aerospace Center (DLR)
👥 Collaborators: Mehul Bhatt, Jakob Suchan, Paul Hemeren

This research investigated how visuospatial attributes in narrative films direct viewers’ attention, using eye-tracking (32 participants/scene) and a novel visuospatial event model for semantic annotation. We analyzed 10 film scenes from the Moving Images dataset to correlate attentional measures with multimodal scene features.

Aim

To establish a method for semantically grounding events and examine how visuospatial cues (gaze, motion, spatial relations) influence:

  • Attentional measures including synchorny (shared gaze patterns)
  • Predictive (anticipatory) viewing behaviors

Methodology

  • Dataset: 10 scenes from acclaimed films (Solaris, Goodfellas, Grand Budapest Hotel etc.)
  • Participants: 32 viewers per scene tracked at 60Hz using Tobii X2-60
  • Annotation Framework: Expert annotations using ELAN tool with controlled vocabulary for:
    • Scene Elements: Objects, regions, body parts
    • Scene Structure: Visibility, motion, spatial relations, gaze, actions
    • Visual Attention: Fixations/saccades (low-level) + object-level attention
  • Analysis Metrics:
    • Attentional synchrony: % viewers fixating same region/body part
    • Feature distribution in high/low synchrony segments

Visuospatial Event Model

A formal taxonomy for semantic interpretation of scenes. See example below:

Category Features
Scene Elements Objects, regions, body parts
Scene Structure visible(X), moving_towards(X,Y), gazing_at(X,Y)
Visual Attention attention_on(face(X))

Results

  1. High synchrony correlates with:

    • Isolated characters in frame
    • Specific behavioral cues (e.g., sudden gaze shifts)
    • Visibility changes (e.g., occlusions)

  2. Low synchrony segments contained 32% more event data (scene structure annotations), suggesting richer interpretative possibilities when attention diverges

  3. Key attention drivers:

    • Head movements (predictive of upcoming actions)
    • Gaze transitions between agents
    • Hand actions with referential significance

Applications

  • AI Systems: Grounding for human-activity recognition
  • Directing Practices: Quantifying attention-guiding techniques
  • Extended Reality: Predictive gaze models for VR narrative
  • Cognitive Modeling: Benchmark for human-like event understanding
  • Attention prediction models using visuospatial features

Project Outcomes

📚 See Publication

Collaboration Opportunities

Open to collaboration or discussion on methodology, data, or future directions. Happy to exchange ideas and explore new perspectives.