↔ Interactive setup schematic · epipolar imaging connection

Two clocks. Three places to change the measurement.

The source scans slowly across the scene. Within every exposure, much faster intensity modulation encodes path delay. A spatial phase code and a moving shutter do different jobs.

Scan synchronization selects source–sensor correspondences. Modulation phase measures delay. The spatial carrier supports the Fourier reconstruction used in this project.

Gold: illumination. Mint: direct return. Violet: indirect paths. Blue: exposed sensor region. The white outline marks the inspected pixel.
Traces show the selected pixel’s full-scan signal, not a second optical interference experiment. Change a gate or phase code to change the actual computed measurement.
—indirect amplitude admitted
—multipath phase bias
—mean received × reference
—addressed scan positions

What the model computes — and what the animation does not imply

One direct path and three explicit indirect returns connect a 28 × 12 perspective-sampled relief surface to a camera. They include a different column, the same column at another row, and an extra-length return to the same pixel. The weights are specified for explanation; visibility, BRDFs, shot noise and absolute optical power are not simulated.

P = Σⱼ AⱼGⱼ exp(iφⱼ), φⱼ = 2πfLⱼ/c + βⱼ
R(θ) = Σⱼ AⱼGⱼ cos(θ−φⱼ), S(θ) = cos(θ+α)
⟨R S⟩ = ½ Re[P exp(iα)]

These are AC components after removing constant offsets from nonnegative emitted intensity and sensor response. They describe MHz intensity modulation and lock-in correlation. They do not represent interference of coherent optical fields. The waveform animation slows the modulation clock independently of the scan clock; their displayed speeds are not a physical timing ratio.

Point and line projectors share a collocated, calibrated camera geometry. Their image-coordinate phase code is therefore the same for direct paths. A point scan addresses N×M positions, a line scan N columns; this count is not an acquisition-speed or efficiency benchmark. The signal is normalized over a complete scan. A pixel gate is modeled only for a raster point source; a line with a column gate still admits paths within the active column.

For scanned sources, G is a Gaussian in source-to-receiver column offset; the point/pixel pairing additionally gates row offset. It selects the endpoints’ correspondence, not every intermediate bounce. Even a path that leaves the plane can return to an accepted pixel. For steady flood or plane-front illumination, G=1 in the integrated comparison: a moving exposure alone cannot identify which source region launched a photon. The blue display is the instantaneous exposure, while the computed amplitude is integrated over the full scan.

A world-coded source uses β=2πd(½+X/10), a calibrated projector uses β=2πd·x/(N−1), and sensor coding uses α=2πd·x/(N−1). Plane-front tilt changes illumination path length to Z cosθ+X sinθ. A spatial carrier is separate from rolling exposure; one need not imply the other. A calibrated offset projector would require a different correspondence model.

Achar et al., Epipolar Time-of-Flight Imaging ↗ describes scanning an epipolar sheet while exposing corresponding sensor pixels. The selected-transport and AMCW principles motivate this schematic; it is not a reproduction of that hardware.