Estimating 3D spatiotemporal point of regard: a device evaluation

Area of Science:

• Human-Computer Interaction
• Computer Vision
• Visual Neuroscience

Background:

• Current 2D eye-tracking methods oversimplify scene depth, missing crucial visual cues like accommodation and vergence.
• Accurate 3D point of regard (PoR) estimation is vital for understanding visual attention in naturalistic settings.

Purpose of the Study:

• To develop and validate a system for recording spatiotemporal scene information and the point of regard (PoR) in 3D ecological environments.
• To enhance existing 2D visual attention models by incorporating 3D spatial data.

Main Methods:

• Combined head-mounted eye-tracking with a miniature time-of-flight camera for 3D PoR estimation.
• Assessed calibration accuracy by matching PoR to known target locations at varying depths.
• Measured accuracy using omnidirectional mean average error (OMAE).

Main Results:

• The system achieved an overall 3D PoR accuracy of approximately 2° OMAE.
• Accuracy variation over a 1-hour recording was maintained within 3.6° OMAE.
• Demonstrated continuous determination of accommodation and vergence cues in habitual environments.

Conclusions:

• The developed system provides accurate 3D PoR estimation in ecological settings.
• This method offers a significant advancement for visual attention research and understanding human-computer interaction.
• Enables continuous analysis of visual cues in everyday applications, such as mobile device usage.