Vision software dashboard: camera monitoring for every application

One screen carrying every inspection station — live image, module status, license state and the parameters behind each verdict — is what a vision software dashboard buys you. The catch sits on the device side, not on the screen: a dashboard can only monitor what the cameras and the controller expose. In practice that means GenICam-compliant cameras on Ethernet and one controller that owns every acquisition session. This page sets out what the monitoring layer must deliver, what it costs in network traffic, where the streams should meet the dashboard, and the checks to run on each camera before you standardize on a single interface. The acquisition engine underneath is its own design problem, covered in multi-camera acquisition session architecture explained.

What must one dashboard show for every vision application?

Four panes, no more: a live image stream with its per-frame metadata, a status line per module (running, idle, fault), the license state of every application, and write access to the parameters that matter — exposure, gain, trigger mode, processing thresholds.

The read path and the write path are separate features, and you should demand both. Pixels prove the optics and the lighting; metadata — timestamp, exposure, gain — proves the acquisition settings, which is how you tell a drifting light from a failing camera without saving a single image. Status comes from the controller's session state, not from the camera: a camera can stream perfectly while its application sits faulted on a trigger timeout. And every verdict has to trace back to the exact frame that produced it, a timestamp discipline covered in matching vision verdicts to frames with timestamps.

Which cameras can a central dashboard actually see?

Every GenICam-compliant camera with an Ethernet interface. GigE and 10GigE area scan cameras stream their pixels natively. Smart cameras and vision sensors report over Fast Ethernet or GigE. USB3 cameras appear only through the host that owns their USB bus.

GenICam defines the feature names a dashboard reads and writes — exposure time, gain, trigger source — and GigE Vision adds discovery, so a compliant camera answers any compliant software. The dashboard host needs the camera family's GenTL producer installed, one per interface type rather than one per camera. Three facts to confirm on the data sheet before you buy: compliance stated as GenICam and GigE Vision (or USB3 Vision), a fixed-IP mode, and PoE or a power plan that matches your wiring.

Smart cameras do their processing on the device, so the dashboard monitors verdicts and the last image instead of a pixel stream. A 5 MP color vision sensor such as the MV-SC3050XC-16M-ATN-H talks Fast Ethernet and nothing else — it joins the dashboard as a network device, not as a managed camera. Bringing the first GigE device onto a controller is covered step by step in connecting a GenICam camera to an industrial controller over Ethernet.

How many live previews fit on one network?

Budget 1 Mbit/s per megapixel: a quarter-resolution preview at 2 fps in 8-bit mono costs about 1 Mbit/s for every MP of sensor. A GigE uplink held at 60% load therefore carries roughly 600 camera-megapixels — 50 cameras at 12 MP, or 300 at 2 MP.

Full rate first, because it dictates the topology. A 6 MP mono camera at 25 fps moves 6,000,000 × 1 × 25 = 150 MB/s, which is 1,200 Mbit/s — more than one GigE link carries. That is why each camera gets its own port on the controller and the dashboard never taps that link. Now decimate: one frame in five at quarter resolution (a quarter in each axis divides the pixel count by 16), and 150 ÷ 5 ÷ 16 = 1.9 MB/s ≈ 15 Mbit/s per camera. Forty such previews total 600 Mbit/s — the 60% ceiling to design to on a 1,000 Mbit/s monitoring link, leaving headroom for polling and reconfiguration.

The LS-CMR-0053 is a 12 MP GigE camera with a 9 fps ceiling. Run flat out in 8-bit mono it moves 12,000,000 × 1 × 9 ≈ 108 MB/s ≈ 860 Mbit/s — nearly a whole GigE link for one camera — while its quarter-resolution 2 fps preview costs about 12 Mbit/s.

The figures scale linearly with resolution and preview rate, so the table holds for any mix:

Color previews in RGB triple these figures. Keep the preview in Bayer or mono, and let the viewing PC render color once, for the single pane an operator actually watches.

Where should the streams meet the dashboard?

On a vision controller that owns the cameras, with one LAN port per acquisition stream and a separate port or VLAN for the monitoring uplink. 5+ LAN ports is the most common layout in the category — 150 of 190 models — so buy that, not a two-port box.

A vision controller rather than an office desktop, for three reasons that all show on the dashboard itself. It is fanless and cabinet-rated. Its PoE models power cameras through the same ports they acquire on — 61 models in the range. And its built-in light channels put lighting health on the same status pane as camera health. The MV-VC3301H-128G66 gives you 5+ LAN ports, 4 built-in light channels and 8 GB of memory. The MV-VC5525-256G70-NN does the same job with PoE camera ports, 8 built-in channels and 32 GB. The MV-SVC2506-16G60-128G separates the networks in hardware: 4 camera GigE ports, 1 configuration GigE port and 1 communication GigE port, so the dashboard and the PLC never share a jack with the cameras.

Three controllers, three network layouts:

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Bind each camera to its port deterministically before you add the monitoring layer — assigning cameras to ports reliably in a multi-camera system. When verdicts also have to land in a PLC, plan that hand-off with the checks in GenICam camera on a PLC controller: what to check. Two special cases: CoaXPress cameras join an Ethernet dashboard only through the frame grabber host that re-publishes their frames. If you are backhauling several feeds over coax, combining multiple camera feeds on one coaxial cable covers the wiring. When the dashboard is really the plant HMI, running one HMI vision project across multiple cells is the pattern to copy.

How do you keep monitoring traffic from slowing inspection?

By decoupling the two paths: the dashboard never taps a camera's acquisition link. It subscribes to the decimated copy sized above, which the controller publishes on a second interface or a VLAN reserved for monitoring.

Decimate and encode at the controller, where the frames already sit in memory; re-encoding forty streams at a viewing PC burns a core for nothing. Metadata-only mode belongs on status walls: timestamp, exposure, gain and verdict per frame arrive in kilobits, not megabits. Reserve moving images for the pane an operator is actually watching.

When previews stutter or frames go missing, treat it as overload and work the diagnosis in image acquisition stalls: diagnose overload from multiple sources. Building the preview pane into your own software? Camera live preview in GUI applications covers the threading.

How do you tune parameters from the dashboard without stopping the line?

Through the same GenICam feature interface the acquisition session already uses. The dashboard writes exposure, gain, ROI or a processing threshold between frames, and the next frame's metadata carries the applied values back, so you confirm the change from the image itself.

One rule makes this safe: a single writer. Parameter writes must go through the session that owns the camera — a dashboard writing while a commissioning laptop also writes is a race you lose on the night shift. The mechanics of setting and reading parameters inside a running loop are covered in camera parameter set and get in a capture loop safely. What the camera actually applied is read back per frame, as in per-frame exposure and gain: verifying camera metadata.

What does license awareness prevent?

Unlicensed modules stay invisible to configuration. The dashboard reads each application's license state at startup, shows only what can legally run, and reports the seat count, so a missing license surfaces at commissioning rather than during the first night-shift changeover.

In the common model the viewing layer costs no license and every inspection application carries its own — counted per camera or per running instance, whichever the suite uses. The cameras themselves never need one: GenICam and GigE Vision devices acquire and stream license-free. A license-aware pane buys three operational things: a capability list an operator can trust, because what is greyed out is unlicensed rather than broken; recipes that fail at load with a clear message when they call an unlicensed tool, instead of a cryptic error at runtime; and a visible answer to which stations would use a tool if you licensed it, before you buy the seats.

What should you verify before standardizing on one dashboard?

Seven checks, all testable in an afternoon: GenICam compliance on every device, an IP plan with room to grow, one LAN port per full-rate camera, a preview budget held under 60% of the monitoring uplink, a metadata path for every frame, a license inventory, and a single-writer policy for parameter changes.

• Every camera's data sheet states GenICam and GigE Vision (or USB3 Vision) compliance — no compliance, no standard dashboard.
• Fixed addresses in one documented range: cameras on the controller's ports, monitoring on its own VLAN or NIC.
• One LAN port per full-rate camera; 5+ ports covers a station cluster with headroom.
• Preview arithmetic summed per monitoring link: at or under 600 Mbit/s on each GigE uplink.
• Per-frame metadata visible for every stream, not only the camera someone complained about.
• License pane populated before the first changeover, not after the first blocked recipe.
• Parameter writes routed through the owning session, one writer at a time, announced to the line.

From here, pick the hardware in the vision controllers category or the camera finder, and let the system configurator walk camera, lens, light and controller in one pass. More integration write-ups sit on the systems topic page. If your monitoring layer must speak a protocol this page has not covered, the contact page reaches the applications team — expert support is part of every order.