Method
How the model works
RefVantage evaluates finite paths from a modeled eye position to sampled points in a selected action region. Simplified body geometry may intersect those paths. The result describes this model — not what an official perceived.
Coordinate system
Meters throughout. x runs along the court length, y is vertical, z runs across the width; the floor is y = 0. The top-down diagram draws world x/z directly. The court drawing is schematic and not regulation-certified; court markings are never used as obstruction geometry.
Body models
Each person is a head sphere (r 0.11 m), a torso box from 47% of height up to the neck, and a leg box below. Boxes rotate about the vertical axis when a facing angle is annotated. The same shapes are drawn in both views and used for every intersection test. There are no arms, hands or pose.
Target sampling
The selected action is a small rectangular region. Five fixed points are sampled: its center and four corners, placed in the vertical plane facing the selected official.
Finite sightline checks
For each sample, a segment runs from the modeled eye point to the target point. Only body intersections strictly between the two (with a 0.001 m tolerance at each end) count. Objects behind the target are ignored, the viewing official's own body is excluded, and the nearest blocker is reported. A target point inside a body, or a zero-length segment, returns invalid geometry rather than a forced answer.
Position interpolation
Positions are hand-entered keyframes. Between keyframes they are linearly interpolated — modeled, not observed. Outside an object's annotated range the position is treated as missing and the result is insufficient data; nothing is extrapolated.
Counterfactual assumptions
Moving an official creates an alternative position at a single paused instant. Every other body stays where it was annotated. The comparison does not predict how play, or anyone's movement, would change. No position is described as correct or optimal, and none is recommended.
Known limitations
Geometry does not establish gaze, attention, call correctness, or an appropriate officiating position. Head movement, peripheral vision, arms, camera limitations and annotation error all matter and are not modeled. A clear path is not proof something was seen; a modeled obstruction is not proof it could not be.
Data provenance
Every scene declares whether it is a synthetic demonstration, manually annotated video, or a controlled staged recording. All shipped scenes are synthetic and do not depict real players, teams or games. Loaded video stays in the browser and is never analyzed.