You need what an ultrasound monitor or a legacy HMI shows on its screen, but the only access is the DB15 analog VGA connector on the back panel. This guide walks the whole chain: choosing a capture device your vision software can actually open, sizing the interface for 1280×1024 at 20-30 fps, digitizing and recording the same stream at once, and the checks that stop a wrong purchase before it ships. The examples assume a Windows host with DirectShow, but the bandwidth, cable and source-setting rules hold on any OS. Work through it in order and the on-site bring-up takes about an hour.
What must the grabber support so my vision software can open it?
Buy a grabber that ships a WDM capture driver: the device then enumerates as a standard DirectShow source, and any vision library that lists webcams will open it at 1280×1024 without the vendor SDK. The driver model is the whole compatibility question. A WDM device registers with Windows as a video capture source, so every DirectShow-aware application — vision packages, ffmpeg, streaming tools — sees it next to a webcam. A device that only ships a proprietary SDK forces every consumer of the frames through that SDK, and you inherit its version conflicts; why one frame grabber viewer works and another fails dissects that failure mode.
The pre-buy check costs nothing: ask for the supported-mode table and confirm the sheet names a DirectShow-compatible WDM driver, not just a bundled recording application. When the box arrives, open the device in GraphStudioNext before touching your vision project. A graph that renders there means your library will list it too. The equivalent selection logic for composite and S-Video hardware is in analog frame grabber integration with vision software, and a longer walkthrough of matching analog hardware to a vision library is in selecting an analog frame grabber for vision software. If your package truly cannot enumerate DirectShow devices, an ActiveX capture wrapper can bridge the gap — a workaround, and current vision packages make it unnecessary.
How much bandwidth does 1280×1024 at 30 fps need?
At 1280×1024 and 30 fps the stream runs at about 79 MB/s in YUY2 (4:2:2) or 118 MB/s in RGB24, so the interface — not the sensor — decides whether you capture uncompressed. YUY2 is what most WDM capture devices deliver; the host converts it to RGB for processing.
MJPEG moves the same frames at roughly 10-40 Mbit/s of screen content, but costs a decode per frame on the host and can smear fast-changing traces — fine for logging, wrong for motion measurement. Count on two to four frames of delay end to end, about 70-130 ms at 30 fps, from the analog buffer, the transfer and the decode. That is invisible for logging and too slow for closed-loop control.
Cropping cuts the number in two different places. A device-side ROI lowers the 79 MB/s before the cable: crop to an 800×600 image area and the rate falls to 800 × 600 × 2 × 30 = 28.8 MB/s, inside USB 2.0 territory — but only if the grabber's mode table offers sub-frame output. A host-side crop after capture, the idea camera ROI cropping at the source covers for machine-vision cameras, saves processing time and not cable bandwidth.
USB 2.0, USB 3.0 or PCIe — which interface should you buy?
Buy USB 3.0 for laptops and carts and a PCIe card for fixed stations: both carry the uncompressed 79 MB/s stream with headroom, while USB 2.0 tops out near 40 MB/s sustained and forces on-device MJPEG.
| Interface | Wire rate | Sustained in practice | Uncompressed 79 MB/s | Verdict |
|---|---|---|---|---|
| USB 2.0 | 480 Mbit/s | 35-40 MB/s | does not fit | MJPEG, or device-side ROI |
| USB 3.0 | 5 Gbit/s | about 350 MB/s | fits, 4× headroom | default for portable hosts |
| PCIe 2.0 ×1 card | 500 MB/s | about 400 MB/s | fits | fixed stations, most robust |
Desktop hosts often run every external port from one shared USB 3.0 controller, where keyboards, cameras and disks compete for the same bandwidth. A separate host card gives the capture device ports of its own. The MV-GU2204 is a quad-port USB 3.0 host adaptor built for exactly that job, and PC requirements for a frame grabber explains the slot and lane budget the card needs.
- Interface
- USB3
- Ports
- 4 ports
On a laptop, run the full interface chain check before ordering: not every USB-C port carries USB 3.0 data, and docks renegotiate link speed mid-stream. Keep the passive USB 3.0 run to about 3 m — the practical limit at 5 Gbit/s — and go active or optical beyond it; the cables category lists high-flex USB 3.0 leads if your device takes a detachable one. Where the station already carries acquisition cards, resolving frame grabber resource conflicts covers the slot and interrupt planning that bites when one more card goes in.
CoaXPress, Camera Link and GigE cards
58 models · from $189.00How do you set up the source machine's VGA output?
Set the machine's output to 1280×1024 at 60 Hz — a standard VESA timing with a 108 MHz pixel clock — so the grabber samples a 1:1 pixel map and on-screen text stays crisp. A native match means every source pixel lands on one captured pixel. Any other output mode makes the grabber resample and blur small text before your algorithms see it. The 75 Hz variant of the same mode runs a 135 MHz pixel clock and pushes weaker analog front ends out of lock, so hold the machine at 60 Hz.
Geometry deserves one check: 1280×1024 is 5:4, not 4:3, and some capture pipelines render square-pixel sources as 4:3, stretching the image by about 7 percent. Display a circle on the machine and measure it in the captured frame. Equal width and height means the geometry is right.
Keep the analog run short. At a 108 MHz pixel clock, shielded VGA cable holds detail to roughly 3 m; longer runs smear fine text before the grabber samples it, and low-grade leads that omit pins cause intermittent sync. If the machine also offers DVI-D, use it: a passive DVI-D to HDMI adapter into an HDMI capture stick removes the analog sampling step entirely, and the DirectShow path is unchanged.
How do you process and record the same stream at once?
Build one DirectShow graph: the capture filter feeds an Infinite Pin Tee, one branch goes to your processing and the other to the AVI writer, because a WDM device opens for exactly one process. Two applications cannot share the device — the vendor recording tool running beside your vision application fails on day one, with the second process seeing no device at all.
Inside the one graph, use the Infinite Pin Tee and not the Smart Tee: the Smart Tee exists for preview and drops frames on its preview branch whenever the renderer falls behind — exactly the frames a measurement wanted. Capture filter, tee, then a sample grabber or your library's source on one branch and an AVI multiplexer with file writer on the other, started by a single run call.
For the file branch, MJPEG or uncompressed AVI reads back in practically every vision package and player. H.264 keeps files three to ten times smaller, but the AVI needs a matching decoder on every machine that reads it, and shipping that decoder with your application can need the codec vendor's license — settle that before a multi-machine deployment. Write the file to its own physical drive; at 10-40 Mbit/s any disk copes, but a shared system SSD shows the contention as stutter.
When your processing cannot hold 30 fps, queue frames between the tee and the algorithm and fix the overflow rule up front: newest-wins for live measurement, all-frames for the record. Fixing the gap when acquisition outruns processing sizes that buffer and its memory cost.
How do you bring the grabber up, step by step?
The whole bring-up is under an hour when the device is right: confirm driver and modes first, then install, enumerate, set the format and soak-test, in that order.
- Confirm on paper: the spec sheet names a DirectShow-compatible WDM driver, and its supported-mode table lists 1280×1024 at your frame rate.
- Set the source machine to 1280×1024 at 60 Hz and connect a short, fully-wired male-to-male VGA cable.
- Install the driver only; close or remove the vendor recording application so it cannot claim the device.
- Open GraphStudioNext, insert the capture filter and render a preview — the machine's screen should appear.
- Open the format negotiation dialog (IAMStreamConfig) and lock 1280×1024, 30 fps, YUY2 — or MJPEG where the bandwidth plan says so.
- In the vision software, enumerate DirectShow devices, open the grabber and set the same format so nothing renegotiates at run time.
- Add the tee, the processing branch and the AVI writer, run the graph, and read the measured frame rate against 30 fps.
- Soak for 30 minutes, watching dropped-frame counters and file growth, then capture an on-screen circle to re-check geometry.
It works when the tool reports 30 fps measured, the dropped-frame counter stays at zero through the soak, and the recorded file plays back inside the same tool. The readback test is the one teams skip and regret at audit time.
What should you check before ordering?
Order once five answers are in hand — a WDM driver, the supported-mode table, the right cable, a free controller or slot, and a codec plan — and ten minutes here stops a wrong purchase.
- Driver: the sheet names DirectShow/WDM support, not only a bundled application.
- Modes: 1280×1024 at 30 fps appears in the supported-mode table, 60 fps if you quantify motion.
- Cable: a fully-wired male-to-male DB15 VGA lead, 3 m or less, plus any DVI-D adapter the source allows.
- Host: a free USB 3.0 controller or slot, checked against the accessories finder for host adaptors.
- Codec: MJPEG or uncompressed unless H.264 readback and codec licensing are confirmed on every reading machine.
- Ownership: one application owns the device; everything else reads files or shared memory.
- Display: a way to hide patient banners on the source before capture where the data is regulated.
When is a VGA capture stick the wrong answer?
Skip the grabber when the machine can export over the network, or when your loop cannot tolerate 70-130 ms: DICOM export and digital I/O each beat screen capture at their own game. Many ultrasound machines push DICOM stills and clips to a server. Analysis that runs on saved images then needs no capture hardware and raises no leakage-current question. When timing closes the loop, low-latency screen capture: digital input or camera weighs a camera on the monitor against tapping the panel signals — both beat a 2-4 frame capture pipeline. And when the HMI cannot output a standard VESA mode, no grabber locks reliably; that project needs the digital route from the start.
For the hardware around the capture device — host adaptors, cables, cards for the cameras that follow — run the configurator from the accessories step, or contact us with the machine's outputs and your target frame rate and we will specify the chain.
The configurator matches a camera, lens, light, light controller and cables to each other, step by step, and lists only parts that fit what you already picked.
Questions
Does a VGA capture stick need its own power supply?
Usually no. USB sticks draw bus power: USB 2.0 supplies 500 mA and USB 3.0 supplies 900 mA at 5 V, which covers a VGA front end. PCIe cards take their power from the slot. If you add an active or optical USB extension for a long run, that extender normally needs its own supply, so plan an outlet at the cart.
Will the capture include on-screen menus and the patient banner?
Yes. The grabber digitizes everything the machine places on its VGA output: overlays, measurement readouts, patient data and menu pop-ups. If changing overlays would trip your algorithms, hide the banner in the machine's display settings before capture, or mask that region in software. Hiding it at the source also keeps it out of recorded AVI files, which matters where patient data is regulated.
Is 30 fps enough for ultrasound screen capture?
For logging, measurements on frozen frames and reading numbers, yes. The screen redraws at 60 Hz, so 30 fps keeps every other frame; fast-moving traces arrive with half the time samples. If you quantify motion — M-mode traces, valve movement — capture at 60 fps. That pushes you to USB 3.0 or PCIe because the uncompressed rate roughly doubles to about 158 MB/s.
Can the vision tool read the recorded AVI back offline?
Only with the right decoder. Uncompressed and MJPEG AVI files read back in practically every vision package and player. An H.264-encoded AVI needs a matching decoder installed on every reading machine, and redistributing that decoder with your application may need a license from its vendor. Record MJPEG during commissioning and standardize on H.264 only after readback is confirmed everywhere.
The source has DisplayPort instead of VGA — what now?
Use an active DisplayPort-to-HDMI or DisplayPort-to-DVI adapter into an HDMI capture stick; the capture path and driver are unchanged. Passive adapters only work when the source supports dual-mode DisplayPort (DP++), which many medical displays do not. Keep the conversion at the source and the digital run short. Converting at the PC end of a long analog line reintroduces the sampling problems you avoided.
