Cool experiments at the intersection of Computer Vision and Sports β½π
I have long been fascinated by the use of Computer Vision in sports. After all, it is a combination of two things I love. Almost three years ago, I wrote a post on my personal blog in which I triedβββat that time, still using YOLOv3βββto detect and classify basketball players on the court.
FIFA World Cup 2022 has motivated me to revisit this idea. This time I used a combination of YOLOv5 and ByteTrack to track football players on the field. This blog post accompanies the Roboflow video where I talk through how to track players on a football field.
I was watching a FIFA 2022 World Cup match the other day, and one of the things that caught my eye was VAR - Video Assistant Referee, or to be more precise, the part of it responsible for analyzing whether a player was on the offside. I did a little research and found that the system performs pose estimation on multiple cameras at once. I decided to check how difficult it would be to reproduce it at home using two cameras and YOLOv7.
This project explores the use of GPT-4V in sports analytics, specifically in the context of football. Its primary aim was to evaluate whether GPT-4V could effectively distinguish and assign players to teams based on the color of their uniforms. This was achieved through the implementation of several advanced vision prompting techniques.
https://github.com/SkalskiP/sports/assets/26109316/b354b38d-1a12-477d-9283-45059ce12467