You have a camera that outputs RGB and IR, the listing said dual-stream, and the machine in front of you shows one device and, at best, one stream in the dropdown. The phrase covers three different hardware designs, and each exposes its channels to the host differently: as a selectable stream inside one driver, as two device nodes, or as multiple streams under one GigE Vision device. The sections below name the design you have from its behaviour, walk through enabling the second channel in each case, and give you the bandwidth arithmetic that decides the frame rate your link will carry.
What does RGB and IR from one sensor actually mean?
Three hardware designs hide behind the phrase dual-stream: a single imager with an RGB-IR colour filter array, one imager split into two filtered halves, and two separate sensors in one module — and each design exposes its streams to the host differently.
RGB-IR colour filter array. A 4×4 pixel pattern in which about a quarter of the sites are IR-passing pixels instead of colour pixels. One readout, one exposure, and an ISP or your own code demosaics the bands apart. Registration is pixel-exact by construction, the IR channel resolves fewer pixels, and white light leaks IR into the colour channels unless the design applies IR suppression or a switchable IR-cut filter.
Split sensor. One imager, two halves, each under its own filter — dual-band designs put RGB over one half and IR over the other. Timing is shared, and registration holds only along the seam between the halves. The LS-BCM-0059 is a GigE board camera with a split sensor, 1.0 MP and Mono on an M12 mount.
Two sensors in one module. Two imagers behind side-by-side lenses or a beam splitter, packaged as one USB or GigE device. Exposures can run independently, and separate lenses parallax-shift close parts between the two views.
Why does the camera show up as one device instead of two?
The host enumerates devices, not streams: one USB serial number or one GigE Vision device ID appears as a single camera, and the RGB and IR channels sit underneath it as selectable streams, extra interfaces or separate device nodes — however the firmware decides to expose them.
Over USB, a UVC camera normally registers one video function with a list of formats, so an application picks one mode at a time. The RGB and IR channels then appear as selectable modes in the same driver or as a channel picker in the vendor utility; firmware that registers two video functions appears as two cameras and can stream both at once. What UVC compatibility really means covers what the class standard promises and what it leaves to the firmware.
Over GigE Vision, one device can open several streams, each with its own UDP port and packet size. How many streams exist is written into the device's GenICam description, and the acquisition library lists them per device. On an embedded board, a MIPI CSI-2 receiver carries up to four virtual channels per port, which is how one sensor block feeds several image paths to a single driver — CSI virtual channels for multiple sensors on one port walks that case.
How do you get both streams at the same time?
You enable the second channel inside one application — through the stream selector in the vendor utility, a second device node, or a multi-stream acquisition in your GenICam code — and you only get two separate camera entries when the bridge firmware registers two video interfaces.
Bridge-based kits load a vendor stream profile that tells the utility how to initialize the sensor and packetize both channels. Without the profile the utility enumerates nothing; with a single-channel profile you see exactly one stream. The profile must match the camera and the bridge board revision, so if it did not arrive with the kit, request the file for your board version from the vendor's support channel, load it at the prompt and relaunch. A plain UVC camera needs none of this — a UVC camera streams with no config file because the class driver reads the formats from the device itself.
When the camera exposes two device nodes, open each with its own capture handle and negotiate format and frame rate per node. Read the nodes from separate threads, because one slow consumer stalls both. Refer to the nodes by name in your capture configuration: USB enumeration order is not guaranteed between reboots.
Can one link carry both streams at full frame rate?
Two simultaneous 1080p Mono8 streams at 30 fps move 996 Mbps, and a Gigabit link tops out near 950 Mbps in practice — so at 30 fps both channels do not fit on GigE: drop to 25 fps, cut resolution, or move the camera to USB3.
| Stream at 1080p, 30 fps | Bytes per pixel | Data rate |
|---|---|---|
| Mono8 | 1 | 498 Mbps |
| YUV422 | 2 | 995 Mbps |
| RGB8 | 3 | 1,493 Mbps |
A GigE Vision stream fills a 1,000 Mbps link to roughly 950 Mbps once UDP and Ethernet overhead are paid — a little less at 1500-byte MTU, a little more with jumbo frames. Packet size is the exact knob turned in 10GigE camera at only 1 Gbps: fix packet size and NIC setup. USB 3.0 signals at 5 Gbps, carries about 4 Gbps of payload after line coding, and lands near 400 MB/s in practice.
The decisions for the example: two Mono8 channels at 25 fps stay on GigE; RGB8 plus Mono8 at 30 fps totals 1,991 Mbps and needs USB3 or better. When uncompressed does not fit, bridge firmware falls back to MJPEG, which trades quality for bandwidth — the artefacts land exactly where you are measuring on a noisy IR channel, so budget for uncompressed when the measurement matters.
Do RGB and IR share the same exposure?
On a single-imager RGB-IR camera both channels come from the same exposure — identical instant, identical exposure time — so a scene change moves both images together, and only the split-sensor and two-sensor designs can expose the IR path independently, and then only if the firmware supports separate timing.
Shared exposure is usually what you want: fused channels only make sense when both frames describe the same instant. It also constrains you — auto-exposure compromises between a channel starved of light and one that saturates, and strobed IR illumination lands in the RGB frame too. On the split and two-sensor designs, confirm whether the channels are hardware-frame-synced; if they free-run, match frames by timestamp the way multispectral band registration aligns channels for fusion.
What do you actually build RGB and IR inspection with?
Match the design to the requirement: pixel-registered bands in one frame buy the RGB-IR single sensor, independent exposure per band buys a two-sensor module or split sensor, and different optics or fields of view per band buy two cameras on one hardware trigger.
Most builds land on two separate cameras, because each channel keeps its own lens, filter and exposure and nothing is compromised. Put them on separate host controllers or a powered hub per camera, trigger both from one line so frames align, and settle the integration points in the board camera evaluation checklist before you order — the shared-bus limits in multiple camera modules on one host are the usual trap.
For a compact build, board-level parts cover each channel:
- Sensor format
- 1/1.8"
- Lens mount
- C-mount
- Sensor format
- 1/2.9"
- Lens mount
- M12
- Sensor format
- 1/4"
- Lens mount
- M12
The LS-BCM-0003 gives one channel 120 fps over USB3 with a global shutter and a C-mount. Pick a mono sensor for the IR channel — silicon keeps useful response out to roughly 1,000 nm — and fit an IR-pass filter matched to your illuminator wavelength. The MV-CB016-10GM-M-W pairs Mono and global shutter with an M12 mount, and board camera lens mount thread sizes is worth a read before you order its optics.
The colour side of the two-camera rig:
- Sensor format
- 1/2.7"
- Lens mount
- C-mount
- Interface
- USB3
- Resolution
- Up to 3 MP
The LS-BCM-0047 covers colour over USB3 on a C-mount at 1.3 MP. When the rig goes inside your own housing, the MV-CB060-10UM-B is a bare-board 6 MP USB3 module built for exactly that.
| Design | Host sees | Exposure | Band registration |
|---|---|---|---|
| RGB-IR single sensor | One device, channels split in software | Shared | Pixel-exact |
| Split sensor | One device, two halves | Shared unless firmware splits them | Along the seam |
| Two sensors in one module | Two streams or two nodes | Often independent | Calibrated; parallax up close |
| Two cameras plus trigger | Two devices | Independent | Hardware-synced frames |
Why does the dropdown still show only one stream?
Work four checks in order: the stream profile the utility loaded, the channel selector beside the format dropdown, the link's bandwidth budget, and whether another application already owns the device — one of the four explains nearly every one-stream-only symptom.
- The profile. A missing or single-channel variant leaves one stream or none. Request the file matching your camera and bridge board revision from the vendor, load it, relaunch.
- The selector. The second channel often hides in a picker beside the format dropdown, labelled per channel. An older utility build may not have the picker at all — compare your build against the vendor's release notes.
- The bandwidth. A device that cannot negotiate bandwidth for both streams quietly exposes one. Recompute the budget from the table above; if the firmware fell back to MJPEG, an uncompressed attempt has already failed once.
- The owner. A capture device serves one application at a time, and a vendor utility left running blocks your second handle — the failure mode behind depth camera color stream fails in OpenCV. Close the utility, then open from your code.
The decisions, per design: a single-imager RGB-IR camera gives you one stream split in software, so there is no second stream to enable. A bridge kit with a channel selector needs the matching profile loaded, then both channels appear. Two device nodes open as two handles on separate threads — budget the link first. A multi-stream GigE device opens one stream per channel with packet size set on both.
Set the camera side of a new build in the camera configurator, filter by interface and shutter in the camera finder, or browse the area scan cameras and board-level cameras ranges. More integration notes sit on the cameras topic page, and questions about a specific dual-band build go through /contact/.
Questions
How do I check whether my dual-stream camera is two devices or one device with two streams?
On Windows, open Device Manager and look under Cameras: two entries means the bridge registered two video interfaces, and you can open both as separate capture handles. One entry means the channels are selectable inside the driver or the vendor utility, not separate devices. On Linux, ls /dev/video* prints the raw nodes, and one UVC camera often registers a capture node plus a metadata node, so test by opening each index before you assume two channels.
Should I light the IR channel at 850 nm or 940 nm?
For silicon sensors, 850 nm typically returns about twice the response per watt that 940 nm does, because sensor sensitivity falls steadily toward 1,100 nm. Pick 850 nm for short range or low power; pick 940 nm where the faint red glow of an 850 nm array would give the system away, and add illuminator power to compensate.
Are the RGB and IR frames automatically aligned for pixel-level fusion?
Only on a single imager. A two-sensor module shifts near objects by its lens separation: at 0.1 mm per pixel, a 25 mm baseline displaces a close feature by roughly 250 pixels between channels. Calibrate the mapping and resample one channel onto the other, as multispectral band registration does, and fix the working distance because the shift grows as the part moves closer.
Can I capture both streams in OpenCV at the same time?
Yes, when the camera registers two device nodes: open two capture handles, one per node, and read them from separate threads so one slow consumer does not stall the other. When the camera is one device with a channel selector, a single handle returns whichever channel is selected, and you switch through the driver or the vendor utility. A USB capture device serves one owner at a time, so close the utility before your code opens the camera.



