Platform
OperaLens is an edge vision layer that runs on your infrastructure, watches the cameras you already own, and turns what they see into counts, events, alerts, and records your operation can act on.
Counting
Each camera is configured with counting lines and polygon zones. From those, the platform continuously computes:
Counting events are debounced — track age, movement thresholds, and per-track cooldowns prevent jittery detections from double-counting.
# current people in the building
BuildingOccupancy = in.1 + in.2
- out.1 - out.2
# recorded every 15 seconds as time seriesRules & alerts
A rule is a threshold on any metric the system produces — zone occupancy, area headcount, a site-level formula — evaluated continuously against live counters.
{
"name": "restricted-zone-3",
"metric": "restrictedZone3",
"camera_id": 4,
"threshold": 1,
"message": "Person detected inside
Restricted Zone 3"
}Setup
The calibration tool grabs a live frame from each camera and lets you draw counting lines and zones directly on it. The geometry becomes the camera’s configuration — no pixel guessing.
Records & reports
Metrics are flushed as timestamped records to the storage you choose — CSV or SQLite with zero setup, MySQL with a pre-built Grafana dashboard, MQTT or webhooks for integration, plus a Prometheus endpoint for monitoring stacks.
Detection classes are configuration: people counting works out of the box, and the same zones, rules, and alerts apply to any class the configured model detects.
timestamp metric cam value 13:41:45 in 4 212 13:41:45 out 4 198 13:41:45 restrictedZone3 4 0 13:42:07 restrictedZone3 4 1 13:42:07 BuildingOccupancy — 37
Start from the problem
One vision layer, configured around what each environment needs to monitor. Capabilities that depend on a specific model are labeled that way.
Define a zone over a hazard area — a press, a robot cell, a loading path. When a person is detected inside, an event fires immediately to the safety channel, an andon system, or both. The event says a person entered Restricted Zone 3 at 09:31 — never who.
Live headcounts per area, entries and exits per line, dwell time in configured zones. Crowded handoffs, empty stations, and shift-change bottlenecks become measured data instead of walking impressions.
With an appropriate detection model, the same zones and rules track forklifts, vehicles, and floor objects: loading-zone occupancy, traffic through pedestrian areas, equipment idling where it shouldn't.
Where a PPE model trained for your equipment and conditions is deployed, the same workflow applies: detect the event class, evaluate the zone rule, alert, and record. It is a configured deployment, not an out-of-the-box toggle.
Alerts handle the moment; records reveal the pattern. Shift summaries turn the event stream into peak occupancy, flow totals, zone events, and alert counts — comparable across days, lines, and sites.
Visitors per hour, busiest weekdays, dwell per zone, and dead spots on the floor plan. See it on real pilot data in the Live Demo.
A pilot targets one camera and one operational problem — and produces one measurable outcome.