Tune Intensity and Point Drops#

The Lidar generic core model lets you control reported intensity separately from the detection threshold that removes weak returns. Choose an intensity processing mode and output scale, then calibrate the detection threshold, beam and detector against the device you are modeling. Attribute definitions are on USD Attributes.

Before you start, configure a lidar that produces points; see Your First Lidar Sensor or the model overview.

Note

The Lidar generic wpm model has intensity controls of its own under omni:sensor:LidarWpm:*, with different names, modes and scale; refer to its USD attributes. Everything on this page is about the Lidar generic core model and its omni:sensor:Core:* attributes.

Intensity controls#

The example sets intensity scale, processing mode and detection-threshold calibration on the sensor prim, using their schema defaults:

USDA

#usda 1.0

def OmniLidar "lidar" (
    prepend apiSchemas = ["OmniSensorGenericLidarCoreAPI"]
)
{
    token omni:sensor:Core:intensityProcessing = "NORMALIZATION"
    float omni:sensor:Core:intensityScalePercent = 255

    float omni:sensor:Core:minReflectance = 0.1
    float omni:sensor:Core:minReflectionRangeM = 200
}

USD Python

from pxr import Gf, Sdf

lidar = stage.DefinePrim("/lidar", "OmniLidar")
lidar.AddAppliedSchema("OmniSensorGenericLidarCoreAPI")
lidar.CreateAttribute(
    "omni:sensor:Core:intensityProcessing", Sdf.ValueTypeNames.Token, custom=False
).Set("NORMALIZATION")
lidar.CreateAttribute(
    "omni:sensor:Core:intensityScalePercent", Sdf.ValueTypeNames.Float, custom=False
).Set(255.0)
lidar.CreateAttribute(
    "omni:sensor:Core:minReflectance", Sdf.ValueTypeNames.Float, custom=False
).Set(0.1)
lidar.CreateAttribute(
    "omni:sensor:Core:minReflectionRangeM", Sdf.ValueTypeNames.Float, custom=False
).Set(200.0)

How intensity is computed#

The detector integrates received light. Intensity processing transforms this signal, then mapping and scaling determine the output value. The detection threshold is applied before output clipping and scaling.

Scale and mapping#

The model computes an intensity between 0 and 1, which omni:sensor:Core:intensityScalePercent scales for delivery and the omni:sensor:Core:intensityMappingType modes can reshape first; the scale, the modes and their tables are defined under Intensity, detection floor, beam and detector on USD Attributes.

Processing mode#

Choose omni:sensor:Core:intensityProcessing to match the device’s reported quantity. See intensity processing for the calculations:

NORMALIZATION (the default) when the sensor reports a calibrated reflectivity: a surface of a given reflectance reads about the same at any range.
RAW when the sensor reports its detector signal as it is: the same surface reads weaker the farther away it is.
CORRECTION when calibration data shows that the device lifts distant returns by a range-dependent factor rather than normalizing them fully.

GAUSSIAN_BEAM and UNIFORM_BEAM trace a finite footprint. At an object’s edge, one firing can produce returns at two distances. In every processing mode, each return’s intensity is normalized to the portion of the beam returned across all surfaces reached by that firing. A beam crossing an edge can therefore produce two returns rather than a single dimmed one. Coverage changes in discrete steps; Gaussian beams weight the footprint center more strongly. See rayType on USD Attributes for beam coverage and power distribution.

Detection threshold#

The detection threshold is calculated for a target with reflectance minReflectance at distance minReflectionRangeM in meters. Reflectance is a fraction: 0.1 represents a 10% target. Lowering it permits weaker returns. See detection attributes.

Range limits and minDistBetweenEchosM can also reject returns. See range windows on Constrain the Field of View and echo separation on USD Attributes.

Use a published reflectance/range pair, such as 10% at 300 m. If several pairs are available, select one for calibration and retain the others as validation targets, as in the Velodyne tutorial. Preserve the test conditions associated with each claim. The Blickfeld tutorial uses the published point that specifies target reflectance and records conditions the model cannot reproduce.

The effect of minReflectionRangeM depends on the processing mode. Increasing it lowers the threshold under RAW and, with the default beam, CORRECTION. Under NORMALIZATION with IDEALIZED rays, range normalization largely cancels the reference target’s falloff, and increasing minReflectionRangeM raises the threshold slightly. Adjust minReflectance to change sensitivity in that configuration.

Three schematic signal-versus-range curves for a target of the reference reflectance under RAW, CORRECTION and NORMALIZATION, each with the detection floor at minReflectionRangeM and the clip at 1 marked. — How the processed signal of a target of the reference reflectance changes with range under each processing mode. The floor is derived at `minReflectionRangeM`: under `RAW` and `CORRECTION` a longer reference range meets a weaker signal and so lowers the floor, while under `NORMALIZATION` the normalized signal is nearly flat and the floor barely moves. The clip at 1 applies to the output only. Schematic, not to scale.#

Detection precedes output clipping to 1. If the threshold exceeds 1, strong returns can still pass, but all report the maximum output intensity. Narrow beam divergence can produce this condition for beam ray types; high calibrationGain can produce it for any ray type.

Note

Surface response depends on its non-visual material label. See Non-Visual Sensor Materials for configuration.

A return behind a transmissive surface must exceed the detection threshold after transmission losses on both the outward and return paths. Low transmission can prevent detection. The front surface also produces a return, so increase maxReturns to report surfaces behind it; see USD Attributes.

Beam properties#

peakPowerW and pulseTimeNs affect received energy. For beam ray types, pulseTimeNs also affects echo separation; see minDistBetweenEchosM on USD Attributes. focusDistM, Msquared, divergence and beam waist define beam geometry. Intensity processing and detection thresholds use these parameters for every ray type, including IDEALIZED. Units and definitions are on USD Attributes.

The beam waist is the beam’s radius at its focus, in meters. The divergence is the angle at which the beam spreads beyond the focus, authored in degrees per axis. The beam quality factor Msquared (M²) scales divergence relative to an ideal beam of the same waist; 1 is ideal and the default is 1.4.

Divergence or beam waist#

Beam waist and divergence are related, per axis, by divergence = M² · λ / (π · waist). Use the same length unit for wavelength and waist. The result is in radians; convert it to degrees for comparison with divergenceHorDeg and divergenceVerDeg.

If either value in one pair is zero and both values in the other pair are nonzero, the model derives both values of the incomplete pair. The default waists are 0.002 m and divergences are zero, so divergence is derived by default.

To specify divergence, set both angles and set the waists to zero. One zero waist is sufficient to derive both. Setting only one divergence angle while leaving the other zero causes both to be derived from the default waists, replacing the authored angle. At least one pair must be complete for this calculation.

If all four values are nonzero, the model retains them without reconciling them. Intensity processing and IDEALIZED rays use those stored values. GAUSSIAN_BEAM and UNIFORM_BEAM tracing uses divergence and a waist derived from it, which can differ from the stored waist. To keep tracing and intensity consistent, specify one pair and set the other to zero.

A Gaussian beam drawn from the emitter, narrowest at the waist at the focus distance and widening along the divergence angle, with a wider dashed envelope for a beam quality factor above 1 and a note on which pair the model derives from which. — Waist, divergence and beam quality factor describe one beam. The waist is the radius at the focus distance `focusDistM`; beyond it the footprint grows along the divergence angle, and a beam quality factor above 1 widens the far-field spread. Author one pair and the model derives the other through divergence = M² · λ / (π · waist). Schematic, not to scale.#

Detector and mechanical noise#

Use measured or published detector values for aperture, quantum efficiency, pixel pitch (micrometers), calibration gain and bit depth. aspectRatio and pixelPitch determine collecting area. calibrationGain multiplies the signal and can raise the detection threshold above the output clipping level. See USD Attributes for definitions.

Mechanical-noise attributes perturb emitter direction and origin. See mechanical noise for controls, units and defaults.

Verification#

Read Intensity over [0, Counts[0]) and check for a nonzero value. An empty output cannot satisfy this check. Count valid points using the Flags bit 0x40, including when includeInvalidPoints is enabled. See Outputs for channel definitions, Configure Lidar Output for invalid-point handling and Sensor Outputs for bounds on array reads.

Then move the detection floor and re-run against the same scene: raise minReflectance and the valid count falls, lower it and the count rises, the dimmest and most distant surfaces changing first. How many points a sweep produces is scene-dependent and not reproducible; the direction of the change is.

Two renders of the same lidar sweep over the asphalt plane and the concrete cube, colored by Intensity. Left, at the default floor, the rings reach the horizon; right, with minReflectance raised to 0.5, the far rings are gone while the cube keeps every point. — The same sweep colored by `Intensity`, on the left at the default floor (`minReflectance` 0.1 at 200 m), on the right after raising `minReflectance` to 0.5. The distant asphalt rings disappear first; the cube, bright and near, keeps all of its points.#

Potential issues#

See the Symptom Index for additional causes and checks.

No returns from surfaces you know are within range: the detection floor is above them. minReflectance is too high, or, under RAW (and CORRECTION with the default beam), minReflectionRangeM is shorter than the distances in the scene. Under the default NORMALIZATION with IDEALIZED rays the range barely moves the floor, so lower minReflectance instead. Refer to the calibration pair.
Intensities outside the expected numeric range: check intensityScalePercent and the processing mode before changing beam or detector parameters.
Every kept return reports the same, largest intensity (intensityScalePercent / 100 under LINEAR mapping): every kept signal reached the clip at 1, as it must when the detection floor itself is above 1. With a beam rayType a very narrow divergence can put the floor there, and a high detector gain can with any ray type. intensityScalePercent only rescales the clipped value; look at the beam and detector numbers and at the calibration pair.
Intensity changed after a change that “only” set the wavelength: expected. waveLengthNm feeds the photon conversion.
Beam-derived quantities not what you computed by hand: both the divergence and the waist were non-zero, so neither was derived. Authoring only the divergence is enough for this, because the waists default to non-zero. Refer to divergence and beam waist.
Two close surfaces arriving as one return: they are closer together than the separation the ray type resolves. Refer to minDistBetweenEchosM on USD Attributes.