Structured-light and laser-snapshot 3D cameras deliver a full point cloud from a single pose, on one condition their mounting guides all print: the camera must be completely still during acquisition. Between acquisitions you may move it anywhere — bolted to a column, riding a robot wrist, carried on a tripod. The mounting decision is not whether the camera may move, but how you reach stillness at the instant of capture and what that stillness costs per cycle. Below is the motion budget in millimetres, the arithmetic behind it, and the mount to pick for each case.
The rule: move any time except during acquisition
The mounting instructions for structured-light 3D cameras say the same thing: the camera must be completely still during acquisition to deliver its specified 3D image quality, and outside acquisition it is completely safe to move — including on a robot or any other moving platform. That is geometry, not legal wording. Projector, part and camera must hold one relative pose for the whole pattern sequence, or the solver triangulates against a scene that changed underneath it.
All three mount families are therefore legitimate, and each pays for stillness differently:
- One station, same pose every cycle → a fixed stand. Stillness is structural, and the only per-capture cost is trigger latency.
- Many poses, parts or stations → a robot arm. Stop fully, dwell 0.5–1 s, then trigger.
- Camera relocated between sessions → a tripod or portable stand. Wait 1–2 s after every move.
If you are still choosing the camera type itself, start with industrial 3D camera selection: stereo, ToF or structured light; this page is about holding still whatever you choose.
Why motion smears a point cloud
A structured-light camera projects a coded pattern and photographs it from an offset viewpoint; each matched projector-and-camera pixel pair triangulates one 3D point. To tell one stripe from its neighbours, most cameras project a sequence of patterns, so the acquisition time spans the whole sequence, not one exposure. Move mid-sequence and the geometry changes between patterns: correspondence breaks, and the damage lands in the depth channel as noise, ripples and holes — not as a graceful blur. Structured light 3D cameras: projector pixel mapping to 3D points shows why a mismatched pair lands at the wrong depth outright.
A residual 1 mm/s drift through a 100 ms acquisition displaces every point by 0.1 mm, which is the entire depth spec of a typical industrial snapshot camera. Faster motion does not degrade politely; the solver mismatches stripes and drops points.
The motion budget: hold motion to a fifth of repeatability
Give the mount a firm share of the error budget: keep camera motion during acquisition below one fifth of the camera's depth repeatability. Structured-light snapshot cameras typically specify repeatability in the 0.05–0.1 mm band at their rated distance, and multi-pattern acquisitions run from tens of milliseconds to a few hundred. Take 0.1 mm repeatability and a 100 ms acquisition as the example inputs, and scale linearly to your two numbers.
- Repeatability 0.1 mm → motion budget d = 0.1 mm ÷ 5 = 0.02 mm.
- Acquisition t_acq = 100 ms → maximum camera speed v = d ÷ t_acq = 0.02 mm ÷ 0.1 s = 0.2 mm/s.
- A stopped arm still sways at structural modes near 10 Hz. Sway amplitude that respects 0.2 mm/s: A = v ÷ (2π·10 Hz) = 0.2 ÷ 62.8 = 0.0032 mm ≈ 3 µm.
- Rotation bites hardest. At 600 mm range, a 0.1°/s yaw drift sweeps the imaged points at 600 mm × 0.001745 rad/s = 1.05 mm/s, smearing 0.105 mm per capture — five times the whole budget.
Three micrometres of residual sway is below anything a robot arm guarantees, which is the real lesson of the arithmetic: you cannot buy this budget with stiffness alone. Buy the first decade with a stiff mount, and the last two with time — stop, let the sway decay, then trigger.
Choosing the mount: fixed stand, robot arm, portable
| Mount | Moves between captures | How stillness is reached | Wait per capture | Pick when |
|---|---|---|---|---|
| Fixed column or stand | Never | Structure does the work; nothing on the station moves | Trigger latency only | One station, same pose every cycle |
| Robot wrist or tool mount | Every pose | Full stop, then decay of arm sway | 0.5–1 s dwell | Many poses, parts or stations |
| Tripod or portable stand | By hand, between locations | Level, release, let the sway die | 1–2 s | Trials, audits, relocated scans |
| Line-scan 2.5D gantry | Continuously | Encoder-locked constant speed replaces stillness | Not applicable | Conveyors, webs, moving gantries |
Stationary: pay once, in steel
A fixed stand is the only mount with no per-cycle cost. Place the camera at its rated standoff — 3D camera working distance and positioning guide covers placement and aiming — then lock the station down. A machine vision camera stand such as the LS-MNT-0040, height 700 mm on a 400 × 450 mm base, is the pattern to copy: mass low in the base, camera high, nothing cantilevered that can ring. For the stationary case, a projected structured-light camera such as this one takes the entire cloud in one still pose:
- Type
- Stereo / structured light
Browse camera stands for columns, slide stages and brackets, or filter the stands and brackets finder by type.
On the robot arm: pay per capture
An arm-mounted camera moves freely between poses — that is the point — and stillness becomes a timing discipline. Mount stiff at the flange: robot wrist camera mounts: stiffness, flange fit and vibration covers the bracket and flange choices. Then gate the trigger on true standstill rather than the end of the motion command. Two further snapshot types ride arms the same way:
- Type
- Stereo / structured light
- Type
- Depth / ToF camera
The galvanometer type sweeps its laser pattern from a fixed pose and wants fringe-projector stillness; the ToF type tolerates small motion at coarser depth precision.
Tripod and portable work: pay in patience
A tripod is a legitimate mount for trials, audits and low-volume scans, because movement outside acquisition is safe. Its sway frequency is lower than an arm's and its damping worse, so wait 1–2 s after each move, and keep hands, legs and cables off it during capture.
The settle wait: how long to hold the arm still
Residual sway decays roughly exponentially, A(t) = A0 · e^(−t/τ). Falling from a 1 mm overshoot at the stop to the 3 µm our budget allows is a factor of about 330, so t_settle = τ × ln(330) ≈ 5.8 τ. Take τ = 0.1 s, a stiff arm after a controlled stop, and the wait is 0.58 s. Commit to a 0.5–1 s dwell after the arm reports standstill: low end for short, stiff arms; high end for long reach or payload at full extension.
The trigger source matters as much as the wait. Acquisition must begin on the controller's in-position or standstill output; camera locate snapshot position: why the robot must return first covers the pose-repeatability half of that discipline. When dwell starts eating the cycle, fix focus and exposure first — reduce wrist camera snapshot time: fix focus and exposure — before you touch the settle.
Checks before you call the station production-ready
- Push test. With the live depth image open, nudge the bracket by hand; ripple should die within about half a second. Longer ringing means stiffen first, dwell later.
- Five-capture repeat. Capture an untouched target five times at production tempo and best-fit the clouds; the spread at the part should sit near the repeatability figure, not at multiples of it.
- Trigger audit. Log the signals once: acquisition starts after standstill plus dwell, never on motion complete alone.
- Cable dress. Service loop at the wrist, nothing tugging the camera or the stand during capture.
- Floor check. Forklifts, presses and neighbouring machines shake stands; if the bench test passes but production drifts, isolate the stand from the floor.
- Worst-pose test. Sway grows with extension and payload, so test the longest reach, not the convenient one.
When stillness is the wrong rule
Laser profilers and line-scan 2.5D systems build the cloud from a line scanned across the part, so camera-part motion is required, not forbidden. There the enemy is speed jitter, and an encoder ties capture rate to travel. The MV-HLVS-16K100-L60 line-scan 2.5D system, 16K resolution with a 600 mm emitting area at a 120 kHz sampling rate, is built for that regime:
- Type
- 2.5D inspection
For parts you would otherwise stop on a belt, the trade between a stopped conveyor with a snapshot camera and continuous scanning with a profiler is laid out in 2D or 3D camera for conveyor pick and place: choosing the system. The stillness rule on this page belongs to snapshot capture alone: relative motion zero while the patterns are on the part.
Putting it together
Start from the camera: the 3D sensors range or the camera finder. Fix the mount next — the system configurator walks stand, camera and controller in order. If your cell has a vibration source you cannot remove, bring the floor spectrum and payload to the contact page and the applications team will size the mount and dwell with you.
Questions
Does a heavier 3D camera need a stiffer mount?
Yes, but stiffness buys little if the source of motion is the arm. Camera and bracket ring together at f = (1/2π)·√(k/m): a 2 kg camera on a 1 N/µm bracket rings near 112 Hz, far above the 5–20 Hz sway of a typical arm, so the arm sets the motion you must wait out. A heavier camera lowers that bracket frequency; a stiff mount keeps it high. Stopping the arm removes the excitation entirely.
How do I prove the settle wait is long enough?
Corner it: shorten the dwell until the five-capture repeat test degrades, then set production dwell at twice the shortest dwell that still passed. That margin covers day-to-day variation in load, temperature and wear. If the passing dwell costs more cycle time than you can spare, stiffen the bracket or shorten the arm's approach path before accepting the delay on every part.
Can I trigger the capture while the robot is still settling?
No. Trigger from the controller's in-position or standstill output, never from the end of the motion command: the command completes while the tool can still drift. At 5 mm/s of residual drift, a 100 ms acquisition smears the cloud by 0.5 mm, five times a 0.1 mm depth spec. If the controller exposes only motion-complete, add the 0.5–1 s dwell in the robot program or PLC after it.
Does a ToF camera need the same stillness?
Less of it. A ToF camera builds each depth frame from a single short exposure, so one frame tolerates motion that would wreck a multi-pattern structured-light sequence — at the cost of coarser depth precision. Choose ToF for hand-carried scanning, moving parts and volume checks; choose structured light when you need fine surface detail, and engineer the stillness described above.
Does the rule apply to the part, or only to the camera?
It is relative motion, so both sides count. A part still coasting on a conveyor smears the cloud exactly like a drifting camera. Stop the part and let its fixture settle — a hard stop on light tooling dies out in a few hundred milliseconds — or move to a scan-in-motion profiler with encoder sync. Never capture during the coast after the belt or shuttle command ends.