Computer Vision-Based Pickleball Analytics System

Computer Vision

A system that leverages YOLOv8 and a fine-tuned ResNet50 model to track players and ball in real time—automating shot speed, player positioning, and court analysis.

Published

February 7, 2025

Introduction

This project analyzes pickleball players in video footage to measure shot speed, player movement, and overall court positioning. It leverages YOLOv8 for player and ball detection and incorporates a fine-tuned ResNet50 model to enhance keypoint precision and positioning analysis. By combining these computer vision techniques, this project demonstrates a real-time, in-depth approach to performance analytics and provides a great opportunity to refine machine learning and computer vision skills.

Output Videos

Here is a screenshot from one of the output videos:

pickleball computer vision screenshot

Models Used

  • YOLO v8 for player detection

  • Fine Tuned YOLO for pickle ball detection

  • Court Key point extraction

  • Trained YOLOV5 model

  • Trained pickleball court key point model

Training

  • Pickle ball detetcor with YOLO
  • Pickleball court keypoint with Pytorch

Requirements

  • python3.8
  • ultralytics
  • pytroch
  • pandas
  • numpy
  • opencv