Sports Statistics Using AI and Video Recognition

Client Company

UK sports-betting operator

Solution Description

A POC project that can benefit gaming/sports betting, both platform operators and odds-making teams. The customer needed the solution for football match analysis. The platform gives an additional stream of statistics in near real-time for game analysis purposes using video to provide positional and relative statistics. Its advanced functionality offers performance predictions based on historical data.

Basic version covers	Advanced capabilities are
Player Recognition (team, number, appearance, position, gate), their pace/speed	Player and team performance - current vs. historical (player pace and activity, team position and ball possession)
Player contacts and outputs (tackle, tackle break, foul, turnover)	Player actions with the ball vs. historical performance
Set play actions: throw-in, corner, free kick, goal kick, penalty / start location / end location	Potential Injury Identification: sudden flag changes in pace or performance
Set play results: Initiation / Action Result / Success / Failure	All performance stats relative to grounds (home, new field) and pitch conditions (weather, time of day)
Ball movements, positions, action types and success rates (shot / pass / success / failure)	Modeling of performance predictions (player, team) based on historical data with real-time adjustments

Technology background

Neural networks and machine learning algorithms are the core of this solution:

Video Processing
based on an open-cv library (a classical pipeline for video processing)
Object Detection
a fast open-source solution based on dlib C++ Library
Person Recognition
usage of machine learning models based on Tensorflow framework
Unique Counting
a proprietary hashing algorithm for future re-identification
Demographics
checking is performed via services that provide data on BSSID and its location (eg., GPS data)

Results and Current Status

The POC proved successful, and the applied technologies and algorithms are now being tested for horse-racing analytics and forecasting. Since there is a clear market-movement towards automated video statistics generation, reporting, and data-mining, this technology has several use cases in different fields.