Head, Gaze, or Finger? Comparing Object Selection Techniques in Augmented Reality for People with Low Vision

This study investigates how people with low vision (PLV) perform object selection in augmented reality (AR). While AR offers significant potential for visual assistance, most interaction techniques are designed for sighted users, ignoring the unique challenges faced by those with visual impairments, such as reduced gaze stability, central vision loss, or difficulty tracking cursors. The researchers conducted a mixed-methods study with 20 PLV and 18 sighted controls to evaluate three common selection techniques: head-based, gaze-based, and finger-based pointing, each using a dwell-based confirmation method.

Participants performed target selection tasks in two real-world environments: a stationary scenario (selecting objects on a shelf) and a dynamic scenario (selecting targets while walking). The researchers implemented a custom AR pipeline using a Meta Quest 3 headset and an integrated eye tracker. To ensure accessibility, they developed a specialized gaze calibration interface that allowed for manual offset correction, accommodating users who rely on a preferred retinal locus (PRL) due to central vision loss. Performance was measured through task completion time, selection stability, and qualitative feedback on mental demand and user agency.

• Performance Trade-offs: Gaze-based selection was the fastest for initial pointing in stationary tasks, but head-based selection proved more stable and less mentally demanding overall, particularly for small targets.
• User Preference: While head-based selection was generally preferred for its stability, participants with central vision loss favored finger-based selection, noting that it provided a superior sense of control and agency compared to gaze or head movements.
• Accessibility Barriers: The study identified a compounded negative effect of reduced visual acuity and restricted peripheral fields on both selection efficiency and stability, highlighting the need for adaptive AR interfaces that account for individual visual profiles.