Video Analysis

Read every frame.

Video Analysis indexes the footage you already own - broadcast, CCTV, drone, dashcam, archives and live streams - into eight structured signals per frame. Behavior and objects only - no facial recognition, no biometric watchlists.

  • 8signals per frame
  • < 10 minto read an hour of footage
  • 20×cheaper than manual review
  • Nofacial recognition

Eight signals on every frame

Each frame is read for what's in it, what's happening and where - grounded in the detections, not asserted.

  • Scenes

    Footage segmented into scenes and each one classified, so hours of video become a navigable index.

  • Objects & people

    Objects and people detected and tracked across the frame - by behavior and presence, never by identity.

  • Brands & placement

    Logos and products spotted, with on-screen prominence and the exact duration they're visible.

  • On-screen text

    OCR reads lower-thirds, captions and signage, timecoded to the frame they appear in.

  • Brand safety

    Risk flagged across categories, so unsafe context is caught before it runs next to your spend.

  • Activity

    Actions and events recognized on the timeline - what's happening in the frame, tied to when it happens.

  • Zones

    Lines and areas you define - crossings and dwell logged as timecoded events you can search.

  • Scene description

    A plain-language description of each scene, grounded in the detections above - searchable, not guessed.

What it reads, and how it connects

Bring the footage you already have - on the feeds and stores you already run, no rip-and-replace.

  • Any footage you own

    Broadcast, CCTV, drone, dashcam, archives and live streams - the same pipeline reads them all.

  • On your feeds and stores

    Connect over RTSP and ONVIF for live cameras, or S3, Azure and GCS for the archives you already store.

  • Out to your systems

    Structured events and the full REST API push into SIEM, dispatch and your own tooling - cloud or on-prem.

One hour, fully indexed

A single segment - every frame read into eight signals and written to a searchable index.

A fictional example, shown to demonstrate the pipeline.

Video Analysis · frame indexIndexing

Broadcast archive · 42-min segment · 1080p

Scenestudio interview · two speakers · seated00:12:30
Objects2 people · desk · 1 product on table
Brandslogo on mug · on screen 38s · center frame
Text (OCR)lower-third "LIVE · MARKET UPDATE"00:12:34
Brand safetyno flagged categories
Descriptiongrounded in detections · exported to index
Per hour of footage< 10 min

Read the footage you already own.

Send us a slice of your archive or a live feed - we'll run a two-week pilot and show you every frame, indexed.

Contact sales