Counting what a camera sees, not what it detects

A detection pipeline runs over a stored video, a public URL or an allow-listed camera stream, and its results land three ways: as time-series metrics, as a searchable knowledge object per run, and optionally as one entity per tracked object.

The tracker is the point — nine hundred boxes of one parked truck is one truck, and that is the number that belongs on a dashboard.

How it runs

1. Point

   Define a pipeline and point it at a document, a URL or an allow-listed stream.

   Vision pipelines

2. Detect

   The worker posts detections back incrementally, so a live feed is watchable and a long run survives a crash.

   Tracking

3. Materialize

   Results materialize into metrics, a knowledge object, and optionally tracked entities.

   Metrics

4. Alert

   Alert rules read those metrics like any other telemetry.

   Alert rules

The sampling caveat, stated

Tracking assumes consecutive sampled frames overlap. On 1080p60 footage of a single walking worker, a stride of 30 reported 13 distinct people and a stride of 5 reported 2. Detection counts are unaffected; only the unique-object count degrades. Keep the sampled interval under ~0.2 s when that number matters.

Cameras are private-network by nature, so a stream source is refused unless its prefix is on the operator’s allow-list.