A near-infrared (NIR) vein finder shines light through skin; deoxygenated blood in subcutaneous veins absorbs NIR more strongly than the surrounding tissue, so a camera under the headset sees the veins as darker lines. Attenuation follows the Beer–Lambert law:
I = I₀ · e^(−μ(λ)·d)
contrast ≈ 1 − I/I₀ = 1 − e^(−μ(λ)·d)
where d is vein depth and μ(λ) is tissue absorption, which is lowest near the 760–850 nm "NIR window" and rises toward both 700 nm and 1000 nm — so contrast peaks around 800 nm and fades at the slider extremes.
The headset's NIR camera is not exactly at your eye — it sits a small baseline b away. The overlay it draws is correct from the camera's viewpoint, but your eye sees the real arm from a slightly different angle, so anything not lying exactly on the display surface appears to shift. For a vein at depth d, viewed from distance D, similar triangles give the classic optical-see-through AR parallax registration error:
error ≈ b · d / D (b, d, D in the same units)
This is why real AR surgical/vascular overlays keep the camera as close to the eye as possible and re-render every frame from tracked head pose — a wider baseline or a closer subject makes the green overlay drift further from the true blue vein beneath the skin. The scene visually exaggerates this offset ×20 so it stays visible; the readout above shows the true, un-exaggerated millimetre value.
- Vein depth — deeper veins scatter more NIR light (lower contrast) and amplify parallax error.
- Wavelength — sweep it to find the 800 nm sweet spot real vein-finders use.
- Baseline / viewing distance — headset geometry; a tighter baseline or standing back reduces registration error.
- Detection confidence combines contrast and registration error: a clinician trusts the overlay only when both the signal is strong and the green line still sits on the real vein.