Universal path

FI Tech + Generic ONVIF / RTSP: AI Vision on Your Existing CCTV

FI Tech is camera-agnostic. If your camera or NVR exposes ONVIF Profile S/T or simply publishes an RTSP URL, we can run PPE, intrusion, fall, fire, license-plate and occupancy models on the stream — no specific VMS required. Detections route to your VMS, webhook, syslog, MQTT, email, Telegram or our Riyadh-hosted console. Typical pilot deploys in 1–2 weeks on a Riyadh-hosted, PDPL-aligned data path.

Talk to integrations teamRead setup guide

Compatibility matrix

Supported product linesAny ONVIF-conformant IP camera or NVR — Dahua, Uniview, Vivotek, Lorex, Reolion, TP-Link Tapo, Amcrest, Foscam, Ezviz, Imou, generic OEM. Any device that publishes an RTSP URL.
Required ONVIF profileProfile S (live streaming) is the minimum. Profile T preferred for H.265 + events. Profile M used when the camera publishes analytics objects. Profile G optional for NVR-side playback.
AuthenticationONVIF digest user, RTSP basic / digest, optional 802.1X / TLS client certs, IP allow-listing per stream
Streaming protocolsRTSP/RTP (H.264, H.265, MJPEG), RTSPS, optional WebRTC bridge for browser viewers, MJPEG-over-HTTP for legacy devices
Event channelONVIF metadata (PullPoint or BasicNotification), webhook (HTTP POST), syslog, MQTT, email-bridge for legacy NVRs
FI Tech deployment topologyFI Tech inference VM pulls each RTSP URL on the LAN. Detections are published to whichever sink the customer chooses — VMS, SCADA, BMS, SOC, Telegram or our Riyadh-hosted console.

Setup guide

Seven steps from "verify ONVIF" to "test alert".

  1. 01

    Verify ONVIF profile or RTSP URL

    Confirm the camera or NVR speaks ONVIF Profile S/T (test with the free ONVIF Device Manager) or that an RTSP URL is exposed (typical pattern: rtsp://user:pass@host:554/Streaming/Channels/101 or vendor-specific path).

  2. 02

    Open the LAN path to FI Tech

    Allow RTSP 554/TCP and ONVIF 80 or 443/TCP from the FI Tech inference host to each device. No outbound internet is required for on-prem mode.

  3. 03

    Provision a service account

    Create a dedicated ONVIF / RTSP user on each device (Operator level, separate from admin). Avoid sharing the admin password with the inference host.

  4. 04

    Install the FI Tech connector

    Deploy the FI Tech generic connector container on Ubuntu 22.04 LTS, RHEL 9, Windows Server 2022, or as a Docker Compose stack on a small NUC / Jetson edge box.

  5. 05

    Add cameras and AI models

    In the FI Tech console, paste each RTSP URL or run ONVIF discovery on the subnet. Assign AI models per stream — PPE, intrusion, fall, fire, license plate, occupancy. Save the mapping.

  6. 06

    Configure alert routing

    Pick where detections go — webhook to your VMS / SOC, ONVIF metadata back into a generic NVR, syslog into your SIEM, MQTT into a SCADA / BMS, email or Telegram.

  7. 07

    Test the end-to-end alert path

    Stage a controlled scene. Verify the detection reaches the chosen sink within ~1.5 s and that bounding boxes, snapshot, timestamp and zone metadata are intact.

AI models supported on any ONVIF / RTSP feed

  • PPE compliance

    Hard hat, vest, gloves, boots, harness — any ONVIF camera with usable framing.

  • Intrusion / perimeter

    Polygon zones, loitering, line crossing for any RTSP feed.

  • Fall detection

    Slip and from-height events on industrial, retail and elderly-care streams.

  • Fire and smoke

    Early-stage detection day and night IR, on any optical or thermal RTSP source.

  • License plate (LPR)

    Saudi + GCC plates from any RTSP feed positioned for plate capture.

  • Occupancy and people-counting

    Per-zone counts, queue length, dwell time across mixed-vendor camera estates.

Sample latency and accuracy

End-to-end latency
0.8 – 2.0 s end-to-end (camera → inference → chosen sink)
Throughput
10 – 25 FPS per stream depending on model and GPU class
Model accuracy
PPE 92–96% mAP, intrusion 96%+ recall, LPR 94%+ on Saudi plates

Typical figures — depend on camera resolution, codec, lighting, GPU (T4 / L4 / A2000), and network jitter on long RTSP paths.

Saudi-specific deployment

For KSA customers, FI Tech offers a PDPL-aligned data path: raw video stays on the customer LAN or NVR, inference runs on a sidecar appliance on your LAN or in our Riyadh region, and only structured detections leave the camera network. Alert routing is hosted in Riyadh with a documented residency boundary, and audit logs are exportable for NCA / SDAIA review.

Frequently asked questions

We have mixed cameras from many vendors — does this still work?

Yes. The ONVIF / RTSP path is exactly designed for mixed estates. We have onboarded sites with 5+ vendors per fence line and unified them under one detection schema.

Our cameras are old and only speak RTSP — no ONVIF. Can FI Tech still work?

Yes. RTSP alone is enough for inference. ONVIF is preferred for discovery and event injection back into a VMS, but if it is missing we fall back to webhook / syslog / MQTT for alert routing.

What about cameras behind NAT or on a 4G / Starlink uplink?

We support pull-mode at the edge (a small Jetson / NUC on-site running the connector) with detections forwarded over a TLS tunnel. Only structured detections traverse the WAN, not raw video.

How do we know if a camera is ONVIF Profile S/T conformant?

Check the ONVIF conformant-products database (onvif.org/conformant-products) or run the free ONVIF Device Manager against the camera. If neither works, point at the RTSP URL directly and we run analytics anyway.

How does data residency work in KSA?

Raw video stays on the customer LAN or NVR. Inference runs on-prem or in our Riyadh region; only structured detections cross the network. Audit logs are exportable for NCA / SDAIA review.

Ready to add AI to any camera?

Send us your RTSP URLs or run an ONVIF discovery — we will scope the rest on a 30-minute call.

Contact FI Tech