Skip to content
A bank of stadium floodlights at night, rows of lamps in a grid against the dark.

OTT & Media · AI in production

Highlights cut while the match is still running

A model that watches the live feed, finds the moments worth replaying, and cuts the compilation before the second half.

Client
A major MAM and media logistics company
Segment
OTT & Media, sports
Content
Live streams and video on demand
Timeline
A working real-time app after 2 months
  • YouTube upload
  • Vimeo upload
  • Dailymotion output

Before

What was true before

Highlights were assembled by hand after the match. Each compilation waited on a person to watch the footage, find the moments and cut them, which put the output hours behind the event and capped how much soccer the business could monetise.

After

What we did

A model watches the live stream, recognises more than 15 kinds of on-pitch activity, identifies players by face and by kit, and assembles the compilation as the match runs. Replay detection picks the moments worth a second look.

Build log

How it was built

work log · OTT & Media · Media logistics company7 steps
  1. 01

    Built an ML engine that decomposes a match and compiles highlights automatically

  2. 02

    Detected fouls, red cards, scoreboards, audience reactions and other replay-worthy moments

  3. 03

    Identified players by face and by kit for player-specific scene search

  4. 04

    Added replay detection that skips irrelevant scenes and ads

  5. 05

    Processed live streams with low delay, and video on demand

  6. 06

    Added manual filters for activities and players, and direct upload to video platforms

  7. 07

    Found ad insertion spots and made the system trainable for new highlight types

Measured

Results

As reported in the published case.

reduction in video processing time
72%
accuracy identifying scoring moments and fouls
90%
faster highlight compilation
3.2x
increase in footage processing speed
70%
kinds of on-pitch activity recognised
15+
of delay on a live stream, at its lowest
10 seconds
The published case on oxagile.com

Engagement

Where this work sits

Next step

Yours would be the next one.

Tell us where it breaks, and you get a written answer on what breaks first and what it costs.