Depth camera performance vs operational temperature range

A depth camera datasheet prints two temperature ratings in the same table, and they answer different questions. The performance range is where the vendor still guarantees everything the datasheet promises: depth accuracy, working-distance envelope, field of view. The operational range only promises the camera will not fail. If you deploy a camera that measures — profile, volume, robot guidance — signing off from the wrong row buys accuracy you will not miss until parts start rejecting. This page decodes both ratings, works a 0-35 °C datasheet against a 38 °C line, and ends with the design rule for each case.

Two ratings, two different promises

The performance row is a guarantee about the measurements. Inside it, the published depth error, Z-resolution, working-distance envelope, field of view and frame rate hold as printed. The operational row is a guarantee about the hardware: run the camera in that band and the vendor expects no failure and no damage. Nothing in the operational row says a single published spec still holds.

Between the two rows sits a band where the camera streams happily while its accuracy wanders. Drifted depth maps, warped volume readings and artifacts along depth edges live there. The decision follows: a streaming camera that measures is designed from the performance row, and the operational row only answers whether the hardware survives the deployment. Profile sensors, stereo cameras and time-of-flight units across the 3D sensor range print the table the same way.

Both numbers are ambient air, not the case

The datasheet temperatures describe ambient air at the mounting point — not the housing metal and not the sensor's case. Where a vendor adds limits for internal component cases, those appear as a separate footnote under the table. The distinction matters because a camera's case always runs above the air around it while the electronics dissipate heat, and the gap between core and housing is exactly what the core temperature versus housing temperature method estimates and tracks.

Inside a sealed housing, the ambient that counts is the air inside the enclosure, which sits above plant air by the enclosure's temperature rise. A logger clipped outside the cabinet door reads plant air and tells you nothing about the air your camera sees. Log at the bracket.

Worked example: judging a 0-35 °C rating against a 38 °C line

Inputs, and where each comes from:

• Performance conditions of one widely used stereo depth camera family: 0-35 °C ambient.
• Operational conditions of the same family: up to 40 °C ambient, paired with a humidity rating of 90% RH at 40 °C.
• Your line: a logger at the camera bracket reads a 38 °C average across the hottest hour of the hottest shift. That measured value is illustrative; the two datasheet rows are the vendor's printed numbers.

The rule in words: compare the logged ambient against the performance ceiling first, because that decides whether datasheet accuracy applies; the operational ceiling decides only survival.

Arithmetic for the 38 °C bracket:

• Margin to the performance ceiling: 35 - 38 = -3 °C, so the bracket sits 3 °C above the guaranteed-performance band.
• Margin to the operational ceiling: 40 - 38 = +2 °C, so the camera runs with 2 °C of survival margin.
• Width of the unguaranteed band: 40 - 35 = 5 °C — the stretch where "it operates" and "it measures to spec" disagree.

Result: the camera streams, no published accuracy applies, and a clogged filter or a stalled cabinet fan consumes the remaining 2 °C.

Decision: the design target is bracket ambient at or below 35 °C. Either move the camera or add airflow until the logged worst case sits under the ceiling with margin, or run a temperature-sweep drift test and write your own tolerance band from the results. Above 40 °C the question stops being accuracy and becomes survival; the camera or the thermal design is wrong for the cell.

What drifts first above the performance ceiling

Depth scale moves with geometry. A stereo camera computes distance as z = f × B / d, with focal length f in pixels, baseline B between the two sensors and disparity d. Warm the camera and B grows: aluminum mounts expand about 23 ppm per °C, so a 100 mm baseline running 20 °C above its calibration temperature stretches by 23 ppm × 100 mm × 20 = 0.046 mm, a 0.046% scale shift — about 0.5 mm at a 1,000 mm stand-off, with the world reading slightly farther than it is. That chain is physics, not a datasheet claim: a steel mount moves less, a long cantilevered bracket more.

Active emitters drift with temperature too: projected patterns and pulsed illumination change output and alignment, which shows up first as artifacts along depth edges. Cameras that watch their internal temperature and compensate were calibrated to do so inside the performance band; outside it you are extrapolating the compensation model. Focus moves as lens barrels expand — the focus drift guide gives the diagnostic path, and infrared systems answer the same physics with lens athermalization. Factory calibration was taken near room temperature, and the calibration and recalibration guide covers when the drift justifies recalibrating.

Which cameras carry this table

Every depth family in the shop prints the same structure, and the row that matters follows the job. A laser profile sensor bolted beside a hot press fails by accuracy long before it fails electrically, because its whole product is repeatable profile geometry.

Projected structured-light cameras for volume and pick cells live in wider spaces — logistics halls, pallet corners — where the seasonal swing moves the bracket ambient far from room temperature across a year.

Laser galvanometer stereo cameras sweep larger fields from a fixed stand-off, and the same two rows govern them; only the weighting of the drift mechanisms changes.

Thermal cameras: the same split plus a scene number

Thermal and infrared cameras carry an ambient window for measurement accuracy the same way, and they add a scene-side performance figure: the effective temperature range over which the image keeps usable contrast, a metric formalized in NIST's measurement work on thermal imaging camera performance. Read the two separately — the ambient band says where accuracy holds, the scene range says what the image can resolve.

The LS-IRC-0044 is a 640 x 512 LWIR thermal camera on GigE with an 88.5° × 73.2° field of view, and its datasheet takes the same two-question reading as any depth camera's: where do the specifications hold, and where does the camera merely survive.

Design rules for a hot deployment

Design from the performance row for any camera that measures. Log the bracket ambient across the worst shift of the year — summer peak, cabinet closed, fans and filters in their worst state — and hold that logged value at least 5 °C below the performance ceiling as a working margin. The operational row then does its only job: confirming the hardware survives transport, storage on the line and the failed-fan case.

Thermal layout decides which row you can use. Forced air holds the bracket close to plant ambient; a sealed housing adds its own rise and cannot cool below the air you give it. The overheating and mounting checks guide walks the mounting-side fixes, and the zero-service-visit design guide treats the failed fan as a design input rather than an afterthought.

Checks before ordering

• Both rows located in the datasheet, with the performance row as the design number and the operational row as survival only.
• Units confirmed as ambient air at the mounting point, with case limits read as the separate footnote they are.
• Worst-case bracket ambient logged across the hottest shift of the year, cabinet closed, housing rise included.
• Logged value at least 5 °C under the performance ceiling: order on the datasheet numbers.
• Logged value between the two rows: a drift sweep or thermal rework before sign-off.
• Logged value above the operational ceiling: a different camera or a different thermal design.
• Humidity pairing read alongside the temperature rows, with the housing checked for cold-start condensation.

The mechanical half of the sign-off is positioning inside the working-distance envelope; the working distance and positioning guide carries those numbers.

When the usual answer is wrong

"It operates to 40 °C, so we are fine" is the classic wrong turn: the camera streams, nothing errors out, and depth accuracy drifts until pallets mis-stack. The operational rating is not permission. The opposite mistake is reading a wider operational band as a better camera; it describes survival, not measurement, and says nothing about accuracy. And do not transfer one family's rows to another model: read the table of the camera you are actually buying, because the ceilings move between families and generations.

If the line runs hot, send us the logged bracket temperatures with the task tolerance and we will help you pick the camera and the thermal layout — the contact page is the direct route. The camera finder filters the depth families by type and interface, and more depth-camera material sits on the 3D camera topic page.