GazeTrend: Trend-Based Gaze Input for Moving Target Selection in Virtual Reality

In gaze-based virtual reality (VR), selecting unpredictably moving targets in dense scenes remains challenging because crowding and occlusions can interrupt confirmation. We present GazeTrend, a real-time technique that infers selection intent from natural gaze trends rather than requiring precise continuous tracking. GazeTrend ranks candidates by testing whether their angular distance to the user’s gaze consistently decreases, then uses an angle-based scoring model with exponential decay so recent gaze evidence contributes more. Selection is committed through two activation modes: Continuous Lock-On enables selection without precisely tracking the target over an extended period, while Go Back in Time creates a stationary virtual copy at an earlier target position for retrospective selection. We conducted a user study with moving targets and compared GazeTrend with dwell and two advanced techniques. GazeTrend was 72.3% faster than dwell with Continuous Lock-On and 60.8% faster with Go Back in Time.