Missing points in a 3D camera point cloud: fix the depth holes

A 3D camera that produced a dense point cloud at commissioning and now shows ragged holes is one of the most common support cases we see. The camera is rarely broken. Depth from projected structured light or stereo matching depends on the projector's pattern or on surface texture being visible to the sensor, and when the light in the scene drifts, the matcher silently returns nothing for those pixels. This guide walks you through the likely causes in order, the test that isolates each one, and the settings that restore a dense cloud. The examples assume a projected-pattern or stereo 3D sensor, but the same logic applies to any depth camera.

Symptom: holes appear where depth used to be

The pattern is recognizable: the depth image looks moth-eaten, with gaps that grow over days or weeks, often over one part of the scene rather than everywhere. The color image stays normal, and the camera keeps streaming. That combination already tells you the sensor, cable and interface are working, because a bandwidth or power problem produces a black image instead.

Two other failure modes are easy to rule out first. Missing depth only along object edges, or on fuzzy and thin parts, is a physics limit of triangulation, not a lighting problem — see close-range depth dropout on fuzzy surfaces and thin parts. Depth that is present but wrong, tilted or warped points at a calibration problem instead, as in stereo disparity map troubleshooting after rectification.

A healthy depth image of a matte target at the nominal working distance should read roughly 95 to 99 percent valid pixels inside its region of interest. When the number drifts into the 80s over weeks, you have a slow optical cause, not a random fault.

Cause 1: the scene got darker

This is the most frequent cause, and it explains the "it worked for a month" pattern. As seasons change, a skylight or window delivers a fraction of its summer light; a failed high-bay lamp drops a wall into shadow; a neighboring machine's enclosure door now blocks a fixture. The projector's dots or lines are still emitted at the same power, but the contrast between pattern and background falls, and below a threshold the matcher stops returning depth there.

Test: capture a depth image at your normal settings, then illuminate the weak area with a bright, broad-spectrum work light and capture again. If the holes close, ambient light is your problem.

Fix: add controlled scene lighting, ideally one that does not swamp the projector's wavelength, and shade the scene from windows. A stable, owned light source beats a variable one you do not control — the same principle behind stereo depth lighting optimization for dark surfaces.

Cause 2: projector or laser power is set too low

Structured-light and active-stereo cameras expose a projector power or emitter power control, usually as a dimensionless setting with a vendor-documented range. Installers sometimes leave it at a mid or eco value to save power; months later, a slightly dimmer scene pushes the pattern below the visibility threshold and holes appear.

Test: raise the emitter power one step at a time and watch the valid-depth pixel count on a fixed target. If coverage climbs with each step, you have headroom.

Fix: set the power high enough that the depth image is dense at the darkest hour of the machine's duty cycle, then leave a margin of one or two steps. Running the emitter at the top of its documented range for the sake of coverage is a legitimate setting, not abuse — but if you need maximum power just to see a white wall, fix the lighting or the geometry first. Emitter output can also decline slowly over years of use, which is one more reason to log depth density rather than trust a fixed setting forever.

For a fixed-installation alternative that does not depend on a built-in emitter at all, a projected structured-light unit such as the MV-DPS200P-02 or a laser-galvanometer stereo camera like the MV-DLS300P-04 gives you explicit control over the projection source.

Cause 3: exposure and gain are fighting the pattern

Depth sensors have an exposure control for the wavelength band they match in. Too short an exposure and the pattern is faint; too long, or with gain pushed up, the image saturates or becomes noisy and the matcher again returns nothing. Auto-exposure, meant for the color stream, can also drift the depth band without you noticing.

Test: lock exposure to a fixed value, capture at several settings, and plot valid-depth pixels against exposure. The curve usually has a clear plateau; you want to sit in the middle of it, not at a cliff edge.

Fix: disable auto-exposure for the depth stream, set a manual exposure inside the plateau, and only then adjust emitter power. If the camera rejects an exposure value, the causes are covered in color stream exposure control: diagnose rejected settings and in GenICam exposure feature discovery for missing attributes.

Cause 4: the surface changed

Texture-less and specular regions drop out of any depth image. A freshly painted wall, a polished chute, shrink wrap, or a part that gained a glossy coating reflects the pattern away from the sensor instead of scattering it back.

Test: aim the camera at a matte reference target such as paper or an uncoated board at the same distance. Dense depth there with holes on your real surface means the material, not the camera, is the limit.

Fix: change the surface angle, add a thin matte coating or spray on test parts, or tilt the camera by 10 to 20 degrees so specular reflection misses the sensor. For dark, low-texture parts, follow the lighting approaches in our stereo depth lighting optimization for dark surfaces guide.

Cause 5: geometry drifted

Mounting brackets creep, and a camera that slowly rotated away from its commissioned aim changes both working distance and incidence angle on the wall. Depth coverage shrinks near the edges of the depth image field of view, so a small aim change can eat a band of coverage that looks like random holes. Even a rotation of a few degrees at a 2 m working distance shifts the pattern footprint enough to matter.

Test: photograph the installation and compare it with the commissioning photo; mark where the housing points relative to a fixed reference in the scene.

Fix: re-shim and re-tighten the bracket, then re-verify depth against a known flat target. Check for a tilted cloud afterwards, because an aim change after a knock can also disturb the extrinsics — the checks in why a laser triangulation point cloud is tilted cover that.

Choosing hardware when the environment cannot be tamed

Sometimes the right answer is a different sensing principle rather than endless tuning:

A compact ToF unit such as the MV-DT01WMU sidesteps pattern visibility entirely, which is worth considering when a wall's finish or the room's light keeps changing.

To compare the whole range by type, use the 3D sensor finder or start from the beginning in the system configurator.

A verification routine that catches regressions

Whatever camera you run, put one number on the wall: the fraction of valid depth pixels over a fixed region of interest on a fixed target, logged daily.

1. Mount a matte target at the nominal working distance.
2. Capture one depth frame with the production settings, exposure and emitter power locked.
3. Count valid-depth pixels in a fixed rectangle; store the percentage with a timestamp.
4. Alarm if the value falls more than 3 percentage points below the commissioning baseline — for example below 95 percent when commissioning measured 98 percent.

That single trend line separates a drifting environment from a degrading emitter from a loosening bracket faster than any amount of staring at screenshots. Repeat the measurement at the brightest and darkest hour of the duty cycle, because a reading taken only at noon hides a lighting problem. If you also need to reason about where the holes are spatially, storing the data as described in storing point cloud data in a voxel grid for 3D inspection makes the gaps easy to locate and measure.

Prevention checklist

• Lock emitter power and depth exposure to manual values inside their plateau — auto modes drift with the scene.
• Own your lighting: shade windows, use fixtures you control, and check them on a maintenance schedule.
• Log valid-depth percentage daily on a fixed target and alarm on a drop.
• Photograph the installation at commissioning so bracket creep is provable later.
• Re-verify depth after any change to the surface finish, nearby machinery or room lighting.

When the usual answer is wrong

Not every hole is a lighting problem. Very close range, thin edges and translucent materials drop out for geometric and optical reasons no emitter setting can fix; see close-range depth dropout on fuzzy surfaces and thin parts. If holes appeared the day new software was installed, suspect the stream instead — an unrecognized format can silently kill the depth output, as in unknown pixel format: diagnose missing depth streams. And if the holes are confined to one side of the image from day one, that is field-of-view clipping, covered in depth image field of view: why stereo coverage is smaller.

If the trend line keeps falling with everything else verified, the emitter may be aging. Measure against a second identical camera if you have one, and route the unit through the manufacturer's documented service path rather than pushing power beyond its documented range. For help choosing a 3D sensor that fits a difficult lighting environment, use the 3D sensor finder or contact our applications team with a depth screenshot and the scene's lighting conditions.