ToF Depth Calibration: Beyond the Datasheet

Time-of-Flight sensors promise simple depth measurement: emit light, measure return time, compute distance. Reality is far messier.

Sources of ToF Error

Systematic errors (predictable, correctable):

• Phase non-linearity in demodulation
• Integration time effects
• Temperature-dependent delays
• Lens distortion

Scene-dependent errors (harder to correct):

• Multi-path interference (light bouncing off multiple surfaces)
• Mixed pixels at depth edges
• Reflectivity-dependent bias
• Motion artifacts

Random errors (noise floor):

• Photon shot noise
• Sensor read noise
• Quantization

The Calibration Challenge

We need to characterize and correct systematic errors while understanding the limits imposed by scene-dependent and random errors.

Distance Calibration

Use a flat target at known distances (0.5m, 1m, 2m, 3m, 4m, 5m). At each distance:

1. Measure reported depth across the FOV
2. Compare to ground truth
3. Fit correction polynomial or lookup table

But: correction depends on integration time, modulation frequency, and temperature. Full characterization requires a 4D calibration space.

Temperature Calibration

ToF sensors drift with temperature - electronic delays change, and optical components shift.

Our approach:

1. Calibrate at multiple temperatures (15°C, 25°C, 35°C, 45°C)
2. Fit thermal model for each calibration parameter
3. Monitor device temperature in operation
4. Apply temperature-compensated correction

Reflectivity Compensation

Low-reflectivity surfaces (dark materials) return less light, causing:

• Higher noise
• Potential bias (fewer photons = different systematic errors)

We're building a calibration target set with controlled reflectivity patches (5%, 20%, 50%, 90%) to characterize this relationship.

Multi-Path Mitigation

The hardest problem. When light bounces off multiple surfaces before returning to the sensor, the measured phase is a weighted average - neither distance is correct.

Approaches we're exploring:

• Multi-frequency ToF (different wavelengths separate direct/indirect)
• Spatial coding (structured illumination patterns)
• Learning-based correction (train on sim data with known multi-path)

None are perfect. Multi-path may remain a fundamental limitation.

Calibration Data Management

Each device has hundreds of calibration parameters. Managing this:

• Unique device ID → calibration record
• Versioned calibration algorithms
• Traceability from field issue to factory data

Building this infrastructure now saves enormous pain later.