Comparison of Visual Saliency for Dynamic Point Clouds: Task-free vs. Task-dependent.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Virtual Reality

Background:

• Understanding human visual attention is crucial for effective human-computer interaction.
• Dynamic point clouds in virtual reality present unique challenges for visual attention research.
• Existing datasets often lack task-free conditions, limiting the study of task-dependent attention.

Purpose of the Study:

• Introduce the Task-Free eye-tracking dataset for Dynamic Point Clouds (TF-DPC).
• Investigate the influence of high-level tasks on human visual attention in VR.
• Compare visual saliency maps from task-free and task-dependent experiments.

Main Methods:

• Collected eye gaze and head movement data from 24 participants in a VR environment.
• Participants observed 19 scanned dynamic point clouds with 6 degrees of freedom.
• Utilized Pearson correlation and an adapted Earth Mover's Distance to compare saliency maps.

Main Results:

• Qualitative and quantitative analyses showed significant differences in visual attention based on task influence.
• Visual saliency maps differed notably between task-free and task-dependent conditions.
• Gaze and movement trajectories provided insights into attention for dynamic point clouds, especially human figures.

Conclusions:

• High-level tasks significantly impact visual attention in dynamic point cloud observation within VR.
• The TF-DPC dataset offers valuable data for studying task-dependent visual attention.
• Findings guide the development of improved visual saliency models and VR perception systems.