Solutions
Computer Vision Infrastructure
A vision deployment is cameras, a GPU node and a great deal of storage, tied together by a network that has to carry all of it. Uniqcli quotes and integrates the whole chain; the analytics model stays yours or your software vendor's.
- Category
- Vision capture, inference and retention infrastructure
- We quote
- Cameras, PoE switching, GPU inference nodes and retention storage
- Boundary
- We supply the chain; the analytics model and its decisions stay yours
The camera count is the input to every other number
Vision projects are usually scoped by cameras and then surprised by everything downstream. Each stream has a bitrate; multiply by camera count and you have a network requirement. Multiply by retention days and you have a storage requirement that is almost always larger than the first estimate. Decode and inference for those streams sets the accelerator requirement, and every camera needs power delivered over the same cable carrying its video. Uniqcli quotes that chain as one system — capture, transport, inference and retention — and integrates it before delivery, so the parts are known to agree with each other.
Four stages, each sized by the one before it
Capture sets everything. Resolution, frame rate and codec determine the per-stream bitrate; sensor and lens choice determine whether the model has anything usable to work with in the lighting the site actually has. A camera specified for a brochure image and deployed into a backlit doorway produces expensive noise.
Transport is where budgets quietly break. The switching has to carry aggregate bitrate without contention and deliver PoE budget across every port simultaneously — a switch rated for PoE on some ports is not the same as one rated for all of them under load. Cable runs, uplink capacity and the pathway between the camera field and the head end all belong on the same drawing.
Inference and retention close the chain. Decode is not free: many streams spend more accelerator capacity on decoding than on the model itself, which is why the node is sized against stream count and resolution rather than against a model name. Retention is arithmetic — bitrate times cameras times days — and it is the number most often underestimated by an order of magnitude at the pilot stage.
Capture, transport, inference, retention
Quoted as one chain so the parts agree with each other, and screened per line before the quote goes out. Naming a manufacturer describes the market, not a Uniqcli partnership or endorsement.
Cameras and capture
Network and surveillance cameras quoted against the resolution, frame rate and lighting the site presents. Where a project standardizes on a specific camera line, it is sourced on request through authorized US distribution and quoted to that standard.
PoE switching and cabling
Allied Telesis and other switching quoted for full simultaneous PoE budget and aggregate throughput — not a port count that only works when half the cameras are off — with patch panels, structured cabling and the uplink capacity behind them.
GPU inference nodes
NVIDIA accelerators in Lenovo hosts, sized against stream count, decode load and resolution rather than a model name. Quoted with the decode budget stated, because that is where multi-camera deployments actually run out of headroom.
Retention storage
Nexsan arrays and Western Digital capacity drives sized by bitrate times cameras times retention days, with drive shelves and the host path quoted alongside so the array arrives connectable.
What stays yours to run
The analytics itself — the model, its training data, its thresholds, and what the organization does when it fires — is yours or your software vendor's. We do not build detection models, we do not tune them, and we do not make claims about accuracy, false-positive rates or what a system will identify. Those are properties of software and of the environment it runs in, not of hardware we can quote.
Nor do we monitor anything. We do not operate a monitoring center, watch feeds, or run a service on your behalf. Video management software and analytics platforms are sourced on request through authorized US distribution where a design calls for them, and the operation of the system stays with the organization that owns the cameras.
The privacy, retention and access policy around recorded video is a governance question with real legal weight in public-sector deployments, and it belongs to your program. We size the storage to whatever retention you have decided on and say nothing about what that decision should be.
Related capability and infrastructure
Vision infrastructure questions
How much storage will we need?
It is arithmetic once three numbers are fixed: per-camera bitrate, camera count and retention period. Send those and we will size it precisely. If retention has not been decided yet, we quote a configuration that can grow by adding shelves rather than replacing the array.
Can one GPU node handle all our cameras?
Sometimes, and the limiting factor is usually decode rather than the model. A node that comfortably runs a detection model on ten streams may saturate at forty purely on video decoding. We size against stream count and resolution and state the assumption in the quote.
Do you provide the analytics software?
Video management and analytics platforms are sourced on request through authorized US distribution rather than held in stock, and we quote them as licensing lines when a design calls for them. Selecting, configuring and operating that software stays with your team or your software vendor.
Is this the same as physical security and access control?
No, and it is worth keeping the two apart. Physical access control — readers, credentials, door hardware — is a separate discipline covered by our access capability page. A vision deployment may sit alongside it, but the equipment, the design questions and the vendors are different.
The solutions atlas
Every solution, one accountable partner.
UniQ platforms
By technology
By customer
- TAA & NDAA-889 Compliance Screening
- CMMC & CUI Solutions
- Federal & DoD
- State, Local & Education
- Healthcare
- Enterprise
- Rapid Procurement & GPC Buys
- Multi-Vendor Integration Projects
- eProcurement & Custom Catalogs
- FISMA Modernization
- CJIS-Compliant Justice Cloud & Local AI
- Federal Storage Modernization
- Government ERP & Business Systems Infrastructure
- Managed Procurement
- Secure AV & Conferencing
- Fiber Network Infrastructure
- Satellite & Resilient Connectivity
- Wavelength & Optical Transport
- Decentralized Data Centers
- Data Center Design & Build
Send the camera count and the retention period
We will size the switching, the inference node and the storage against them, and quote the chain as one screened bill of materials.