Polypilot product mascot

Introducing PolyPilot:

Our AI-Powered Mentorship Program

Learn More
Go to Polygence Scholars page
Maggie Du's cover illustration
Polygence Scholar2023
Maggie Du's profile

Maggie Du

Monta Vista High SchoolClass of 2023San Jose, California



  • "How can machine learning networks be improved for object tracking in football to help referees make calls more efficiently during matches?" with mentor Clark (Jan. 25, 2023)

Maggie's Symposium Presentation

Project Portfolio

How can machine learning networks be improved for object tracking in football to help referees make calls more efficiently during matches?

Started Mar. 19, 2022

Abstract or project description

In recent years, the applications of machine learning have been expanding throughout a variety of fields to help humans perform different tasks with greater precision. Specifically, in football, it gradually became incorporated through systems such as goal line technology, which detects whether or not the ball has crossed into the goal, and Video Assistant Refereeing, which intends to help the main referee with making tough decisions on penalty calls, offsides, and more during matches. Without such technologies, the game would be highly prone to human errors and may prove to be either unfair when bad calls are made, or inefficient and slow when refs take extended periods of time to make a decision. By researching different possible machine learning models that can process datasets of videos and/or numbers from football matches, we can find new ways to track the ball and players in moving images more accurately and efficiently. This could potentially assist in making offside calls (when an attacking player is behind the penultimate defender) by tracking the live positions of the players in question, as well as detecting when a goal is scored by tracking the live position of the ball in relation to the goal frames. Using machine learning alongside the human referees would prevent more controversial calls to be made during important matches that have negatively impacted many teams in the past. I plan on beginning my research through existing image datasets and video footage from previous football matches which I can analyze using neural networks to process the images and detect where important objects are. YOLO is a commonly used live object detection algorithm that is a good place to start experimenting, but by diving deeper into different algorithms, we can compare and contrast all the possibilities and perhaps find one that is the most efficient and accurate.