3D visionTroubleshooting

Depth camera queue timeout no frames: isolate a dead sensor

A depth camera enumerated on USB 3 whose queue hits a timeout with no frames has one dead sensor — run the per-sensor test and send the vendor the evidence.

LumaSpec Vision11 min read Share on LinkedIn

On this page 9 sections
  1. 01What does a 'required 3, received 0' queue timeout mean on a depth camera?
  2. 02Is the cable or the port to blame when the camera enumerates on USB 3?
  3. 03How do you test each sensor on its own with debug logging?
  4. 04What does the output look like when one sensor is dead?
  5. 05When is it a hardware fault and not your code?
  6. 06What do you send the vendor for an RMA?
  7. 07What do you replace a stereo depth camera with after a dead sensor?
  8. 08How do you keep the replacement from dying the same way?
  9. 09Questions

Your depth camera enumerates, the viewer or SDK sees it at USB 3 speed, and then the output queue times out with a line like 'required 3, received 0'. The pipeline is not slow — one image sensor never produced a frame. This guide gives you the isolation order that settles it: one arithmetic check to rule out the cable, the vendor's minimal per-sensor capture example with debug logging, the output signature that separates a silent sensor from a starved link, and the evidence pack that gets a hardware return accepted. It is written for anyone integrating a fixed-mount stereo or structured-light 3D camera on a robot cell or production line.

What does a 'required 3, received 0' queue timeout mean on a depth camera?

The pipeline asked its output queue for frames and received none at all before the timeout expired — for example 3 required, 0 received after 15 s — which, with the device enumerated on USB 3, points to a sensor that never started, not to a slow link.

The timeout message is the end of the chain reporting failure, and it tells you which link failed. The host and camera talk over a control path — enumeration, configuration, calibration reads — and the camera answers with frames. 'Received 0' means the control path came up but the data path never delivered a single frame: at least one sensor inside the camera failed to start or produced nothing from the first frame on.

The arithmetic shows how extreme received 0 is. A sensor streaming at 30 FPS delivers a frame every 33 ms (1 ÷ 30 s⁻¹ ≈ 0.033 s), so a request for three frames is satisfied in 0.1 s:

t_fill = required_frames / sensor_fps = 3 / 30 FPS = 0.1 s frames_offered_in_window = sensor_fps × timeout_s = 30 × 15 s = 450 frames

Fifteen seconds of silence at that rate is 450 frames that never existed. No realistic slowdown turns 450 into 0 — a saturated link drops and delays frames, it does not produce perfect silence from one stream. Treat received 0 as a sensor that never streamed and work down the causes in order. The RGB-side twin of the same decision is covered in RGB stream timeout on a depth camera: SDK support vs hardware fault.

Is the cable or the port to blame when the camera enumerates on USB 3?

Probably not this time: enumeration at SuperSpeed proves the control path, and a cable or power fault throttles every stream together, so one silent sensor beside full-rate neighbours is a camera-internal fault more often than a cable fault.

Spend five minutes clearing it anyway. Run the camera from a direct host port with the shortest cable that reaches — no dock, no hub. Hub and dock faults show a different signature: black or frozen windows on every sensor, random disconnects, late frames. Both are covered step by step in Depth camera shows a black window: fix the USB dock and hub path and USB camera keeps disconnecting: isolate cable and hardware faults.

Then check bandwidth with the same arithmetic, because a data wall throttles every stream together:

bandwidth_Bps = width_px × height_px × bytes_per_px × fps × sensors

Illustrative inputs: a stereo pair of 1.3 MP mono sensors at 30 FPS in 8-bit moves about 78 MB/s (1,300,000 px × 1 B × 30 FPS × 2 = 78 MB/s), far under the roughly 400 MB/s a 5 Gbit/s link carries after encoding overhead; even a 4 MP pair stays near 240 MB/s. A saturated link drops frames on every sensor — it never silences one beside full-rate neighbours. If the fault follows the camera to another port, the port is cleared and the per-sensor test comes next.

How do you test each sensor on its own with debug logging?

Run the vendor's minimal capture example — not your pipeline — with the SDK's debug log level switched on, so each image sensor is started and previewed on its own with none of your configuration in the way.

Close every process that could own the camera first — a wrapper, a viewer left open, a script from the last run — because a second owner leaves sensors unstarted and spoils the result. If your host keeps stale settings between runs, reset to a known state first; the procedure is in 3D camera test flaky on leftover settings: start from a known state.

  1. Set the SDK's log-level environment variable to debug before launch, so every sensor-start attempt is printed to the terminal.
  2. Run the vendor's minimal capture example — the SDK sample that opens one preview per image sensor — not your own pipeline.
  3. Let it run for about ten seconds, stop it, and keep the complete terminal output from the first line to the interrupt.

What you want from the run is the start-up sequence: the SDK lists every sensor it enumerates, prints the USB speed, then opens one preview window per sensor with a live frame-rate counter.

What does the output look like when one sensor is dead?

A dead sensor leaves a precise signature: the device enumerates, every sensor appears in the start-up list, the healthy ones open preview windows at full frame rate — 30 FPS in the reference case — and the failed one never opens a window or yields a single frame.

Read the run against this table:

What the run shows What it means Where it leads
All sensors enumerated, healthy ones at full rate, one sensor with no window and no frames Sensor start fault inside the camera Cold boot, direct port, then the RMA decision below
A window opens but stays black or freezes mid-run Projection, exposure or dock problem The dock-and-hub guide linked above
All sensors drop frames or sit below their rated rate Bandwidth, power or an exposure cap Three depth cameras drop to low FPS: test one stream at a time, Depth camera frame rate drops: exposure settings cap the FPS
The wrapper starts cleanly but the image topic stays empty The same silent sensor, seen from middleware Judge the sensor at SDK level before touching the node graph

Two details make the dead-sensor case certain. The calibration data still reads cleanly, because it lives in the device's EEPROM on the control path — a clean dump proves you hold the right unit, not that every sensor streams. And the debug log stops at the same sensor's start line on every run, while the healthy sensors' start lines are each followed by frame-rate output at their rated rate.

Keep the whole log, not the summary line: support reads the start-up sequence line by line, and the interrupt at the end proves the run was stopped deliberately.

When is it a hardware fault and not your code?

Call it hardware after three clean eliminations: a cold boot on a direct port with no second process owning the device still leaves one sensor at zero frames, the rest hold full rate, and a known-good unit runs every sensor on the same cable and port.

Each alternative has already been removed by the sequence itself. Your pipeline is out because the minimal example fails the same way. Cable and port are out because the fault follows the camera in the swap test. Bandwidth is out by the arithmetic above. SDK support is out because a mode the device cannot run is rejected at configuration time with an error, not with silence at frame time.

The decision rule, stated once:

  • One sensor at zero frames after a cold boot, a direct port, no second owner, and a known-good unit clean on the same cable: hardware fault — go to the return.
  • Every sensor below its rated rate on every unit you try: environment fault — power, hub chain or an exposure setting is capping the streams. Stay on the host side.
  • The failing sensor opens a window but intermittently stops: still a hardware fault, but collect two or three runs of evidence so the pattern is undeniable.

What do you send the vendor for an RMA?

Send four things and the replacement decision goes fast: the camera's serial number exactly as the SDK reports it, the full per-sensor test log with debug logging on, the exact queue-timeout message, and your platform details — OS, host controller and SDK version.

The vendor's return process runs on evidence, and those four items are that evidence. Add the enumerated connection speed the camera reported. Do not open the housing: a stereo camera is aligned and calibrated as one assembly, and an opened unit gives up the warranty position you are about to claim.

If the camera came from us, the same bundle works here: every order includes free expert support, the cameras carry a 3-year warranty minimum, and the contact page starts the claim. For a replacement from the 3D sensors range, keep the measured per-sensor result — it is also the acceptance test you run on the new unit on day one.

What do you replace a stereo depth camera with after a dead sensor?

Treat a dead sensor as a whole-unit fault: the stereo pair is calibrated as one assembly, so the replacement is a new calibrated camera rather than a swapped board, and the same per-sensor test then serves as your acceptance check.

Prefer a unit that generates depth on-camera from its own stereo pair, structured-light projector or time-of-flight array, so the host receives finished depth rather than raw sensor streams it must keep alive — that also shrinks the bandwidth budget you checked above.

Options that do this job:

3D sensorsHikrobot MV-DPS200P-02 Projected structured-light 3D cameraMV-DPS200P-02
Type
Stereo / structured light
$8,829.00 · quotedView and quote
3D sensorsHikrobot MV-DLS300P-04 Laser galvanometer stereo cameraMV-DLS300P-04
Type
Stereo / structured light
$16,579.00 · quotedView and quote
3D sensorsHikrobot MV-DT01WMU ToF sensing cameraMV-DT01WMU
Type
Depth / ToF camera
Price on requestView and quote

On the host side, a vision controller removes the hub question entirely, because its LAN ports are built for camera traffic. The MV-VC5525-256G70-NN lists PoE camera ports and 8 built-in light channels, so camera streams and light triggering run from one box. Size the complete rig in the system configurator before you order anything.

Vision controllersHikrobot MV-VC5525-256G70-NN VC5000 Vision Controller, 12th-I5, 32GB+256G SSD, Supported PCIEMV-VC5525-256G70-NN
LAN ports
5+ LAN ports
USB 3 ports
3-4 USB 3 ports
PoE
PoE camera ports
Light outputs
8 built-in channels
$4,479.00 · quotedView and quoteBuild a system with it

How do you keep the replacement from dying the same way?

Most depth-camera electronics fail young or not at all, so the prevention list after a verified sensor death is short: mount the camera rigidly, feed it from a direct host port instead of a hub chain, keep SDK and firmware versions aligned, and keep one owner per device.

Mounting first: the same stillness that protects image quality during capture protects the sensor assembly from vibration fatigue, and the procedure is in 3D camera mounting: stillness during capture decides image quality. Versions second: mismatched builds fail in ways that mimic sensor faults, and aligning them is covered in 3D camera header version mismatch: fix firmware and SDK alignment. Ownership third: one process opens the camera, everything else waits.

Record the healthy baseline while the new unit is on the bench: per-sensor frame rate at the rated setting and the queue read time in milliseconds. The next silent sensor is then proven dead in one ten-second test, and the return goes back with numbers attached. The 3D camera topic page collects the rest of the isolation guides.

Questions

Can a USB hub or dock cause the same queue timeout?

It can, but with a different signature. A dock that starves bandwidth or power shows black or frozen windows on every sensor, random disconnects or rates stuck well below spec; it rarely silences exactly one sensor while its neighbours hold full rate. Run the camera once from a direct host port with a short cable, and if the one-sensor silence survives the move, stop chasing the cable and run the per-sensor test.

How large should the queue timeout be on a healthy stereo camera?

Small values are enough once streaming has started. At 30 FPS a frame arrives every 33 ms, so a three-frame request fills in about 0.1 s; a timeout of a few seconds leaves ample margin for sensor start-up. If a healthy rig still returns zero frames over a 15 s window, the timeout length is not the problem — a sensor produced nothing, and no timeout value fixes that.

Does a readable calibration dump prove the image sensors are healthy?

No. Calibration and board identity live in the device's EEPROM, which the host reads over the control path, and that path works even when one imaging sensor never starts. A clean dump confirms the unit's identity and its lens data; only the per-sensor capture test proves that every sensor actually streams frames. If the dump itself fails, that is a different fault on the storage or control path, and it changes the evidence you collect first.

Can you swap the dead sensor board instead of replacing the camera?

Not as a field repair. A stereo depth camera leaves the factory as one calibrated assembly: each sensor's position relative to the others is measured and stored in the calibration data. A replaced sensor invalidates those relationships and with them the depth accuracy, so vendors replace the calibrated unit whole rather than swap boards in the field.

The wrapper node starts cleanly but no image topic ever publishes — is that the same fault?

Usually, yes. A wrapper can launch, subscribe and report itself ready while the sensor beneath it never streams, because node start-up and frame arrival are separate steps. Put the middleware aside, run the vendor's minimal capture example with debug logging on, and judge the sensor on that output before debugging your node graph or topic names.