Stereo Vision Depth Triangulation

Stereo Rig

Baseline B 12.0 cm

Focal length f 700 px

Pixel quantization noise (±0.5 px) Auto-dolly rig (moving robot)

Selected Target

Object #0 — click any shape in the main view to select it

True depth Z

— m

Disparity d

— px

Estimated depth

— m

Error

—

Depth resolution at this range (±1 px)

— m

How it works

Two cameras with parallel optical axes, separated by a horizontal baseline B, both photograph the same point. Because the cameras sit at different positions, the point lands at a different pixel column in each image — its disparity d. Near objects shift a lot between the two images; far objects barely shift at all.

disparity:  d = x_L − x_R  (pixels)
triangulated depth:  Z = f · B / d
depth-resolution error:  ΔZ ≈ Z² · Δd / (f · B)

Here f is the focal length in pixel units and B the baseline in metres. This simulator renders the scene from a real left camera and a real right camera (both visible as inset views), measures each target's actual on-screen pixel shift, and plugs it back into Z = fB/d to recover a depth estimate — exactly what a robot's stereo-vision pipeline does after matching features between its two lenses.

Baseline — a wider rig gives more parallax per metre of distance, so depth estimates get more precise, at the cost of a larger blind zone up close.
Focal length — a longer lens (higher f) magnifies disparity too, improving precision but narrowing the field of view.
Pixel noise — every real sensor quantizes to whole pixels; toggling it shows why stereo depth error grows with the square of distance (ΔZ ∝ Z²) — the reason LiDAR, not stereo, is preferred for long-range robot perception.

Real-world relevance: this Z = fB/d relation is the exact equation behind every stereo-camera module used in robotics and AR (Intel RealSense, ZED, and the human visual system itself, which triangulates with a ~6.3 cm eye baseline).

LEFT CAM

RIGHT CAM

Drag to orbit · click a shape to select it as the stereo target

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.