MIXI, Inc.


Innovation in the streaming of publicly-managed betting sports races

In June 2020, MIXI launched TIPSTAR, a service that allows users to watch free live streams of auto race and keirin races while betting with their friends online. TIPSTAR utilizes several internally-developed tools and systems to offer live streams from keirin and auto race stadiums nationwide. We will continue to innovate in the world of sports broadcasting in order to provide new and exciting viewing experiences.

Development of a proprietary automatic AI-based editing system for live video of auto race and keirin races

We have developed a system that uses AI to automate the editing of live video from 48 keirin and auto race stadiums across Japan. Our system decides when to overlay text and graphics and plays suitable background music. This type of work is usually done manually in a studio using dedicated equipment, but MIXI has developed new technology (patent pending) for use with TIPSTAR which pioneers an AI-based approach.

How the AI-based live video editing system works

Underlying technology: Python 3 / TensorFlow Lite / Vertex AI / Edge TPU• Uses image and audio recognition AI simultaneously to make inferences about the status of races

  • Edits video by calling the editing API (BreezeCast, described below) according to race status
  • Infrastructure consists of on-premise servers equipped with and optimized for Edge TPUs

Development of BreezeCast, a proprietary, browser-based tool for editing video from keirin and auto race stadiums

TIPSTAR makes use of an original tool, BreezeCast, which allows editing work to be done on a browser. Text, graphics, and background music can be added to video streamed from each race venue. Normally, video editing requires a specialized environment and equipment, but with BreezeCast, video editing can be performed anywhere with an internet connection. We developed BreezeCast in response to the COVID-19 situation making it impractical for a large number of staff to work together in an editing studio.

BreezeCast from a technical perspective

Underlying technology: Vue.js / Go / C++ / Blackmagic DeckLink / WebRTC

  • Infrastructure consisting of GKE and on-premise servers
    • All containerized, including those on-premise
  • Editing application logic written in Go
    • Information on races and racers (text, graphics, music, etc.)
    • Generating images to display the results of races
  • UI for video editing built using Vue.js
  • Drawing process for editing handled using C++
    • Coaxial port SDK drawing process
    • Video editing API manages the coordinates of each resource, manages layers, and makes edits based on the status of the race
Above is the BreezeCast screen which shows the original video (left) and edited video (right) during an actual race. The “AI mood” is displayed below the videos and is what determines which text, graphics, and music will be utilized based on the status of the race.