Real-time tracking of visually attended objects in virtual environments and its application to LOD

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Cognitive Science

Background:

• Interactive virtual environments require efficient methods for tracking user attention.
• Conventional saliency maps (bottom-up) alone are insufficient for accurately predicting visual focus.
• Integrating goal-directed (top-down) information is crucial for understanding user intent.

Purpose of the Study:

• To develop and evaluate a real-time computational framework for tracking visual attention in virtual environments.
• To enhance attention prediction by incorporating both stimulus-driven and goal-directed contextual information.
• To demonstrate the framework's application in adaptive level-of-detail management.

Main Methods:

• A novel framework combining bottom-up saliency maps with top-down contextual information derived from user behavior.
• Implementation on Graphics Processing Units (GPU) for high computational performance.
• User study involving eye-tracking to validate the framework's gaze prediction accuracy.

Main Results:

• The framework achieved high computational performance suitable for interactive applications.
• Accuracy of visual attention prediction was validated against human eye-tracking data.
• The inclusion of top-down context significantly improved prediction accuracy for single and multiple targets.

Conclusions:

• The proposed framework effectively tracks visual attention in virtual environments by integrating diverse contextual cues.
• Top-down information is vital for accurate attention prediction, aligning with cognitive theories.
• The framework offers a hardware-free solution for optimizing virtual environment rendering and user experience.