Uses your cameras
Standard RTSP streams from the IP cameras you already have. Substreams keep processing efficient; counting geometry rescales automatically across resolutions.
Deployment
Built to be deployed, monitored, and maintained.
For IT and OT teams: what OperaLens runs on, how it behaves in production, and how it connects to your systems. Video never has to leave your network.
- Camerasexisting RTSP streams
- Edge nodeon-premise, your hardware
- Rule enginezones · lines · thresholds
- Outputsstorage · dashboards · alerts
Architecture
On-premise by design.
Processes on your hardware
Detection and rule evaluation run on an edge node inside your network. Frames are processed in memory and discarded — no video storage, no cloud dependency at runtime.
Outputs where you choose
Metric records to CSV, SQLite, MySQL, MQTT, or webhooks. A pre-built Grafana dashboard ships with the Docker deployment; Prometheus scraping via the metrics endpoint.
Hardware planning
Published sizing guidance.
Camera capacity depends on model size, stream resolution, and processed frame rate — counting works well at 5–10 FPS per camera on substreams. These are engineering starting points to validate on your scenes, not guarantees.
| Hardware class | Model size | Indicative cameras per node |
|---|---|---|
| CPU only, 8 cores | small | 2–4 |
| Mid-range GPU (e.g. RTX 3060 / T4) | standard | 8–15 |
| High-end GPU (e.g. RTX 4080 / A10) | standard | 20–30 |
| High-end GPU | larger | 10–18 |
Operations
Behaves like infrastructure, because it is.
- Health endpoints — /healthz liveness for Kubernetes, systemd, and Docker probes; /values live counter snapshot; Prometheus-format metrics
- Graceful shutdown — clean stop on SIGTERM; storage flushed and closed
- Stream resilience — camera threads recover automatically from stream failures with backoff
- Release discipline — semantic versioning and a maintained changelog
- Container deployment — Docker Compose stack including database and Grafana dashboard
GET /healthz200
{
"status": "ok",
"uptime_sec": 86211,
"threads": {
"track-1": true,
"track-4": true,
"view-7": true,
"flusher": true
}
}Pilot footprint
What a pilot actually needs.
One camera
An existing RTSP camera covering the area with the operational problem.
One edge node
A machine inside your network that meets the sizing guidance above.
One destination
A dashboard, Slack channel, or webhook endpoint where results and alerts should land.
Bring IT and OT to the first call.
We like hard questions about networks, hardware, and retention — the architecture answers them better than a brochure.