Agents Lookup

Use cases / Decision models (Jev)

Check camera motion frames for people before sending an alert

Ask Jev's image preview whether a person appears in a few frames from each motion event and alert only when one does.

DifficultyIntermediate
Setup time2-4 hours
CategoryDecision models (Jev)

Author's figure: Author's figure: all 23 people caught versus 14 by the older setup, about 200 ms and $0.00008 per event (Todd Dailey). Reported by the builder, not measured by us.

The use case

Homeowners and small shops with cameras get flooded with motion alerts from cars, pets and shadows. Here a few frames from each motion event go to Jev's image preview with a single yes/no question about whether someone is visible. Only events with a yes send an alert to Discord or a phone. The author's older pipeline, which described frames with one model and captioned with another, missed many people. Image input is still a preview, so keep the raw recordings, and do not use this for decisions about identifying anyone.

Example workflow

  1. Trigger on motion events from the camera's NVR or Frigate
  2. Pick three to five frames per event
  3. Ask Jev's image preview a Noul question per frame on whether someone is visible
  4. Send an alert with frames to Discord if any frame is yes
  5. Keep recordings and review missed or false alerts monthly

Tools

Risks and guardrails

Sources

Written by Agents Lookup. Credits and official pages for the tools this use case names: