Operational video intelligence
Count movement and waiting time in footage you are allowed to use.
Your footage holds more than a recording. Find the movement, waiting and changes that deserve a closer look — with the evidence still in view.
Authorized footage. Defined questions. Human review.
Where does movement
become a bottleneck?
Evidence console
Every number on this panel
comes from one recorded run.
Play the licensed recording and the timeline follows it. Click any second to jump there; mark two moments to compare them. What the run measured is shown with its limits; what it did not measure is listed, not estimated.
t = 0 s · tracked detections 1.4 · phase QUIET
Mark two moments to compare them.
Zone “crossing”, whole clip
- Mean tracked occupancy
- 4.95
- Peak (1 s mean)
- 8.8
- Zone entries
- 143
- Closed by lost track
- 102 of 143
- Line crossings in / out
- 63 / 38
- Median track span
- 3 s
Entries by tracked detections, not unique people.
74 % censored: occlusion, leaving view or association failure.
Accuracy unmeasured.
247 track ids are not 247 visitors.
Review window 40–50 s
50 sampled frames · mean 6.52 tracked detections · peak 10. The run's own recommendation: annotate this interval manually before any staffing or capacity decision.
Provenance
- Detector: PekingU/rtdetr_v2_r18vd@5650961749fa+onnx:DmlExecutionProvider
- Association: supervision.ByteTrack (supervision 0.30.5)
- 360 frames processed in 25.87 s wall time (13.916 fps, onnx-directml-gpu)
- Source sha256 3cb900f0f351f371… · export 8ab714aecd398c83…
- Media: ProtoplasmaKid, CC BY-SA 4.0 — globally blurred; the blur is a display change, not certified anonymisation.
Not claimed by this run
- unique people
- independent accuracy
- validated staffing or congestion intervention
- validated forecast
- customer pilot
- production performance or financial cost
- emotions, intent or identity of anyone in the recording
Start with your operation
Start with one location and one question.
Choose a setting to shape a private evaluation. These are proposed use cases; suitability and accuracy need to be measured on authorized footage.
Retail & service
When does a waiting area need a closer review?
Agree a zone and time window, review changes in occupancy, then compare observations with a manual count.
Evaluation output: a reviewed interval, a counting rule and a discrepancy log.
Plan this evaluation ↗Know what you are evaluating
What works today, and what still needs an evaluation
- Available in the local productRecorded footage review
- Processing of authorized recordings, zones and aggregate movement observations, with a report and source provenance.
- Separate controlled workspaceCamera connection workflow
- Requires its own authorization, authenticated workspace and setup. This public page does not connect to cameras.
- Evaluation requiredOperational outcomes
- Accuracy, time saved and workflow improvement must be established on independent annotated clips and your acceptance criteria.
No emotion, intent, theft or guilt inference. A signal can support human review; it cannot establish what a person thinks or why an event happened.
A measured next step
Bring a question; we agree the test before any work.
Start with one location, a recording you are authorized to use and a decision your team needs to make. Agree the deliverables and price in writing before any work begins.
No footage is uploaded here. Keep personal or sensitive information out of the brief.
Discuss the scope with EVY ↗