Patent pending

The parallax depth camera

A single camera, moved in a pure translation, actively creates the motion parallax it measures. Because parallax is inversely proportional to depth, one lens becomes a dense 3D scanner — purpose-built for the 0–50 cm near field.

See the point cloud → The principle

Active parallax

0–50cm: Target working range
1×: Camera — no stereo rig
Dense: Per-pixel depth, not sparse
Passive: No active illumination — no projector or laser

First principles

Depth falls out of the geometry.

Move a camera sideways by a known baseline while it watches a static scene. Every point's image shifts — and the shift is a direct, geometric measure of how far away it is.

depth = \frac{f b}{d_{disp}}

\frac{σ_{Z}}{Z} = \frac{σ_{d}}{d_{disp}} = \frac{σ_{d} p Z}{f b}

f

— focal length

b

— translation baseline

d_{disp}

— measured image disparity

σ_{d}

— feature-tracking noise

p

— pixel pitch

σ_d = 0.5 px · p = 3 µm · f = 3 mm · b = 5 mm

Nearby points sweep across the sensor; distant points barely move. Because disparity scales as 1/depth, the near field — where most sensors are weakest — is exactly where this method is strongest.

One sensor, no baseline rig

The controlled motion is the baseline. No calibrated stereo pair to align, drift, or pay for twice.

Tuned for 0–50 cm

Structured light and wide-baseline stereo degrade up close. The 1/depth relationship makes this method most precise exactly there.

Interpretable & dense

Depth from measured geometry, not learned priors — dense, per-pixel, and explainable.

Watch it capture

The sensor in motion.

A single pass generates the parallax; the reconstruction builds live.

1× playback speed

Reconstruction

Real capture · jet depth · drag to orbit

Scene0°

Roll0°

Viewer H57.7cm

Depth H50cm

Jet

8–34cm

Cull

9–50cm

The output

Dense 3D you can rotate.

The camera returns a full point cloud of the near-field scene. This is an actual capture, coloured by recovered depth with a jet map — red at 8 cm through blue at 34 cm. Drag to orbit it.

◆Per-pixel depth across the frame, not a handful of feature points.
◆Metric & consistent — a known baseline yields absolute scale.
◆Pipeline-ready for manipulation, inspection, and reconstruction.

Request raw sample data →

Near-field 3D, solved

Put one camera where a rig used to go.

Robotic manipulation, inline inspection, biometrics, reconstruction — anywhere close-range depth matters.

Request a Demo →Explore Drone Detection