I recently finished a 6 month internship as an AI Engineer where I started to learn about computer vision with deep learning. Beforehand I finished two Udacity courses (Deep Learning and Computer Vision Nanodegree) in that field. Computer vision captivated me, so that's why I'm starting cv related projects.
Why Lacrosse? I played for several years lacrosse in Germany and simply because of that I'm starting this project to make a real time lacrosse object detection. And I also didn't find any solution on my initial search (Object Detection in sports yes but not for Lacrosse).
This project will be challenging and will take some time that's why I'll also try to use some agile project-management approaches. I'm working alone on this project so there is no need to overdo it but I will use a kanban board to define my tasks in a backlog and track my work progress (I'll use obsidian's kanban plugin). For example that's how I started my board:
MVP
- Detect all players (classes: goalie, shortstick, longpole) and Referees in real time -> Inference on 30 fps videos
Moonshots
- Additionally to the MVP, count and track the players and referees
- Separate the players in their respective teams
- Detect/track the Ball and predict the ball carrier (pose estimation via Cradling the ball?)
Additional objectives
- Detect the goal and crease (circle around the goal)
- Detect the field lines (sidelines, midline, restrainer, GLE, winglines)
- Detect or predict referee's calls/signs (e.g. goal, time-out etc.)
To get a better understanding of the MVP, look at this image:
I'll create the dataset with Roboflow.
Demo GIF Detection to get a short glimpse:
