A depth camera that works all day and vanishes after every reboot is rarely a failed camera. The failure is almost always startup order: the software that drives the camera starts before Windows has finished finding it. This page is for engineers running USB 3D depth cameras on Windows vision controllers or industrial PCs, where a nightly restart is part of the production plan. You will set the camera's services to Automatic (Delayed Start), fix the order in which SDK and firmware get updated, read the three Device Manager entries that tell you which layer failed, and decide when the USB path, not the camera, is at fault. GigE depth cameras get their own note, because network discovery follows different rules. Work through the sections in the order you meet the fault.
Why does my depth camera disappear after every reboot?
In most cases the camera is healthy and the failure is a startup race: Windows starts the camera's driver services before the USB stack has finished enumerating the device, the service finds nothing to bind to, and the camera stays absent until you move those services to Automatic (Delayed Start).
Here is the sequence on a Windows host. Power-up brings up the USB host controllers, each device answers enumeration, and only then does the service manager launch everything marked Automatic. A camera on a busy controller can still be mid-enumeration when those services query for their device. The service finds nothing, reports no camera, and waits for the next boot to try again. Automatic (Delayed Start) holds a service back until the system is idle, about two minutes after boot, which is normally all the margin a USB 3 depth camera needs.
Three signs confirm the diagnosis:
- The camera returns the moment you unplug and replug it, because the fresh enumeration gives the service something to bind to.
- Restarting the camera's service by hand brings the stream back without touching the cable.
- The failure repeats on every boot, not at random, and a warm restart fails exactly like a cold one.
If a replug does not bring the camera back, service order is not your only problem: go straight to the USB checks further down, because that pattern points at power, cable or port. If instead your fault is a stream that misbehaves once the camera is up, the 3D camera topic page collects the capture-side guides.
How do I set the camera services to Automatic (Delayed Start)?
Set every service that belongs to the camera stack to Automatic (Delayed Start) in the Windows Services console, reboot twice and test before changing anything else, because delayed services start about two minutes after boot, after every automatic service, and that wait is what most USB 3 depth cameras need.
- Open the Services console (press Win+R, run services.msc).
- Identify the services the camera's software installed. On desktop depth modules they are typically listed as Sensor Service, Sensor Data Service and Sensor Monitoring Service; on industrial cameras, look for the capture or device service that ships with the camera's SDK.
- Right-click each one, open Properties, set Startup type to Automatic (Delayed Start), and click Start so the service comes up now.
- If the depth module is used for Windows sign-in, also confirm the biometric service is enabled; a disabled one keeps the module unclaimed even when the drivers load.
- Reboot twice. On the second boot, wait at least three minutes before testing, so the delayed services have had their window.
- If the camera is still missing after step 5, run a clean-boot test: run msconfig, choose Selective startup, clear third-party services, reboot and test, then restore Normal startup before the next production run.
What order should SDK and firmware updates follow?
Bring the camera up on the installed versions first, prove it survives two cold reboots, then update the SDK, then flash firmware only with the tool and version the SDK release names — a camera that is not recognized at boot can leave the update stuck in recovery mode.
SDK and firmware are matched pairs: firmware written for a newer SDK reports as no device found on an older one, and that mismatch looks identical to the startup race on screen. Fix the order, not just the versions.
- Verify the camera is recognized after two cold reboots on the versions already installed.
- Update the SDK, reboot twice, and re-verify.
- Flash firmware with the matching tool, reboot twice, and re-verify.
A stream that disappears after a firmware update is a driver-stack problem, not a reason to flash again; rebuild the driver stack in order instead. The camera firmware version mismatch check and the SDK v2 mismatch that reports no device found cover the version pairing, and if the camera lands in firmware-update mode, read the USB ID before reinstalling.
What should Device Manager show when the camera is healthy?
A healthy depth camera shows three separate entries after a cold boot: the camera itself under Cameras or Imaging devices, the vendor's virtual or depth driver entry, and its port under Universal Serial Bus controllers — any entry missing or flagged means that layer, not your application, is the problem.
Read the entries as a ladder from the bottom up:
- No USB entry at all: enumeration never finished. Power, cable, port or a firmware-setup gate — move to the USB section below.
- A USB entry with a warning icon: the device enumerated but the driver failed. Follow the driver-stack install order, drivers before application software.
- Camera and driver entries present but the SDK lists nothing: the application layer is missing its plugin or device filter, and the device-manager-but-not-in-SDK fault explains that split.
After a Windows update, the RGB and virtual drivers of a depth module can vanish while the depth entry survives; reinstall them as a set rather than one at a time. If every entry looks healthy but the capture application still reports no camera, check the operating system's privacy switch for desktop app camera access, and any antivirus that filters device access, before blaming the SDK. If the camera is a desktop depth module that Windows repeatedly refuses to detect, the step-by-step driver detection fix for desktop depth camera modules walks the same ladder with vendor-specific detail.
Could the USB port or hub be the cause?
Yes, and it is the second most common cause after service order: a depth camera wants a USB 3 root port to itself, because a shared hub, a long cable or a port that holds power through the reboot can leave the device unenumerated when boot-time services look for it.
A depth camera streams depth, infrared and colour at once, so it loads the bus harder than a webcam; a hub that looks idle can still be out of bandwidth once all three streams open. Three checks settle it:
- Move the camera to a root port on the controller with a short cable, and reboot twice.
- If it fails only after a shutdown rather than a restart, suspect a port that keeps bus power alive across power-off; that VBUS-always-on enumeration failure after a reboot is a documented pattern, and the fix is a different port.
- Add a powered hub only after the root-port test passes, never as the first fix.
A camera that enumerates but opens with a black or frozen frame usually sits behind a dock or hub splitting the bandwidth; work through the dock and hub path checks, and if the open call itself throws no-matching-device, the USB open failure fix applies.
Does the same fix apply on a Linux vision controller?
The mechanism transfers but the lever changes: on Linux the capture service races the device node instead of a Windows service, so make the capture service wait for the device to appear, and check whether the kernel driver bound the camera to a device node at all.
Containerized capture stacks sharpen the race, because the container engine is another service with its own startup order: if the engine starts before the device appears, the container comes up empty and no amount of replugging the camera helps. Give the capture container an explicit startup dependency on the device being present; the Docker vision service boot-order fix shows how to sequence those dependencies, and running 3D camera control software in Docker covers the rest of the stack.
For a camera that presents as a standard UVC device, check after each reboot that the kernel bound it to a node; the UVC recognition fixes for Linux IoT kernels cover the driver-binding side in detail. A vendor kernel module follows the same rule: it must load before the capture service starts, and that ordering is a startup dependency you set once and keep.
When is it the camera, not the startup order?
Suspect hardware only after delayed start and a direct USB 3 port both fail on a cold boot, with no unknown device appearing in Device Manager and a second PC failing the same way. Then the camera is unpowered, gated off in firmware setup, or locked in firmware-update mode.
| Symptom after a cold boot | Layer it points to | First action |
|---|---|---|
| Camera returns after a replug only | Startup race | Delayed start on the camera services |
| No USB entry at all | Power, cable or setup gate | Root port and short cable; confirm the camera is enabled in the firmware setup screen |
| USB entry with a warning icon | Driver | Reinstall the stack in order |
| Entries healthy, SDK blind | Missing plugin or filter | Install the SDK's device plugin |
| Camera stuck in firmware-update mode | Interrupted or mismatched update | Read the USB ID, then the vendor recovery procedure |
| GigE depth camera absent after boot | Network discovery and boot order | Link-local addressing and startup-order checks |
Some boards gate cameras in firmware setup, so a camera disabled there never appears at any port; that check costs one restart and rules out the whole USB layer. For GigE depth cameras the USB checks do not apply at all, because discovery happens on the network — the GigE camera boot-order fixes live there instead.
Which vision controller keeps a depth camera alive across reboots?
If reboots keep dropping the camera on a desktop tower, move it onto a fanless vision controller: 2 USB 3 ports cover one depth camera plus a trigger device, and PoE camera ports cover GigE depth models.
The arithmetic is the point. A depth camera and its trigger or light line each want their own port, so two USB 3 ports is the floor for a single-camera station, and every port you keep free is margin for the next device. A controller built for the job also enumerates its own ports in a predictable order, which is exactly what the delayed-start services rely on.
The MV-VC2140-128G40-1T carries 2 USB 3 ports, 3-4 LAN ports and 4 built-in light channels, which fits the single-camera arithmetic exactly.
- LAN ports
- 3-4 LAN ports
- USB 3 ports
- 2 USB 3 ports
- Light outputs
- 4 built-in channels
- Memory
- 8 GB
Stations with several cameras or GigE depth models need more: the MV-VC5525-256G70-NN has 5+ LAN ports, PoE camera ports, 3-4 USB 3 ports, 8 built-in light channels and 32 GB of memory.
- LAN ports
- 5+ LAN ports
- USB 3 ports
- 3-4 USB 3 ports
- PoE
- PoE camera ports
- Light outputs
- 8 built-in channels
Vision computers and fanless industrial PCs
190 models · from $1,119.00Match the camera first in the 3D sensors category, the host in vision controllers, or filter by depth technology in the camera finder. If the reboot test still fails after the service, firmware and USB checks above, send the Device Manager screenshot through the contact page and we will read the layer with you.
The decisions, in one line each: replug fixes it, set delayed start; warning icon on a USB entry, reinstall drivers in order; entries healthy but the SDK blind, add the plugin; nothing appears anywhere, work the USB path and the firmware setup gate; GigE camera absent, fix network discovery and boot order — and put any station that reboots nightly on a controller with ports of its own.
Questions
How long does Windows wait before starting a delayed service?
Windows starts Automatic (Delayed Start) services roughly two minutes after the last ordinary automatic service has started, once the system is idle. Build the acceptance test around that: after a reboot, wait at least three minutes before declaring the camera missing, otherwise you will blame the delay setting for a race that was still resolving.
Why does my camera work after I unplug and replug it?
Replugging forces a fresh enumeration that the driver service can bind to immediately, which is why the symptom points at startup order rather than hardware. If a replug always fixes it but a reboot never does, set delayed start on the camera services. If even a replug fails, move on to the USB root-port and cable tests.
Can a BIOS setting hide the depth camera completely?
Yes. Some boards and laptops gate cameras at the firmware level, and a camera disabled there never appears in Device Manager on any port. If no USB video device shows up anywhere after a cold boot, restart into the firmware setup screen and confirm that the camera or USB controller is enabled before replacing anything.
Does a GigE depth camera have this reboot problem too?
It has an equivalent one. A GigE depth camera is not enumerated over USB at boot; it is discovered on the network, so the failure moves to IP setup and boot order, such as link-local addressing, subnet conflicts and service start order. Those discovery checks differ enough that they deserve their own pass after the service fixes.
Should I reinstall the whole SDK when the camera is missing?
Not first. Reinstalling the SDK cannot fix a service that starts before the camera exists, so you reinstall straight into the same race. Set delayed start and verify the Device Manager entries first, and reinstall the stack (drivers before application software) only when an entry carries a warning icon or a driver disappeared after a Windows update.

