The Pipeline: Camera to Stream to Inference.
Vision AI defines what the platform detects. Computer Vision Infrastructure is how a live video feed reaches that detection layer in the first place — the ingestion, stream handling, and camera-agnostic plumbing underneath every capability on this site.
Stream Pipeline
From a Live Camera Feed to Structured Data
Four stages, running continuously, not on a scheduled pull.
Camera
Existing IP cameras, NVR feeds, or mobile device cameras — no proprietary hardware and no teardown of what is already installed.
Stream
Video is pulled over RTSP via MediaMTX, the streaming layer that normalizes feeds from mixed camera vendors into a consistent input.
Inference
Frames are handed to the Vision AI Engine, running on edge or cloud GPU hardware depending on the deployment — detection happens on the stream, not on a stored file.
Structured Output
The result is structured metadata — bounding boxes, classifications, timestamps — ready to route to an alert, a log, or the Command Center.
Real Deployments
Camera-Agnostic Isn't Theoretical — Here's What It Actually Looks Like
The same ingestion pipeline handles six genuinely different physical configurations already running in production.
Fixed IP camera on the drive-thru lane
Vehicle-journey tracking across arrival, order window, and departure — see Store & Kitchen Operations.
Overhead + scan-plane IP cameras per lane
Non-scanned item and ticket-switch detection at the payment plane — see Smart & Cognitive Checkout.
Advidia High-Speed IP Camera (Hanging Chain Line)
Zero-touch attribute extraction as parts pass a fixed inspection bounding box on the conveyor.
Overhead IP + USB Display Mount Camera
One camera digitizes the analog UT graph, the other tracks 360° inspector wheel rotation, simultaneously.
Mobile Device Native Camera
On-device OCR extraction with no fixed camera infrastructure required at all.
Multi-Zone Industrial IP Cameras
24/7 PPE compliance and machine Start/Stop state classification across multiple zones at once.
Camera-Agnostic by Design
The infrastructure is built to ingest whatever is already mounted on the wall — commercial IP cameras, NVR streams, and mobile-native captures for workflows like the LP Unit’s mobile traceability scanning. There is no requirement to standardize on a single camera vendor before deployment can start.
- Standard RTSP / IP camera ingestion via MediaMTX
- NVR and existing CCTV stream compatibility
- Mobile device camera capture for field workflows
Inference runs on GPU hardware — either on-premise at each location, on GPU-equipped edge devices, for the lowest possible latency; or centrally in the cloud, fed by an on-site VMS (Video Management System) that streams camera feeds up for customers who would rather not invest in edge GPU hardware at every location. Which model fits is a deployment decision made per customer, not a fixed architecture.
Whether that inference runs on-premise or in the cloud is a deployment choice — the full breakdown of both models is covered on the Edge AI page.
Connect Your Camera Feeds to Proglint AI
Schedule a customized demonstration with our enterprise AI engineering team. Discover how Proglint deploys across 2,500+ locations with zero camera teardowns.