This repository contains the codebase for a comprehensive computer vision project designed for analyzing tennis matches. Leveraging advanced techniques and models, the system offers multi-faceted insights into gameplay dynamics, player movements, and ball trajectories.
Here's the link to the post on LinkedIn: https://www.linkedin.com/feed/update/urn:li:activity:7184230793005740033/
-
YOLOv8 Player Detection: Utilizes YOLOv8 for robust player detection within tennis match footage, enabling precise tracking of player movements throughout the match.
-
Fine-Tuning for Ball Detection: Fine-tunes YOLO model to accurately detect the tennis ball in varying conditions, ensuring consistent tracking and analysis of ball trajectories.
-
Speed Detection: Implements algorithms to calculate the speed of both players and the ball, providing valuable metrics for performance evaluation and tactical analysis.
-
Mini Court Generation: Dynamically generates a miniature representation of the tennis court, reflecting the actual positions and movements of players during the match for intuitive visualization.
-
Key Point Extraction: Fine-tunes the last layer of ResNet50 on a custom dataset to extract keypoints on the court, facilitating advanced spatial analysis and strategic insights.
-
Installation:
- Clone this repository to your local machine.
- Install the required dependencies using
pip install -r requirements.txt.
-
Running the System:
- Add your tennis match video in the .mp4 format inside the input_video folder.
- Run the main script to perform player detection, ball detection, speed calculation, mini court generation, and key point extraction.
-
Customization:
- Modify the configurations and parameters in the scripts to tailor the analysis to your specific requirements.
- Fine-tune models on additional datasets for improved performance in different conditions.
Contributions, bug fixes, and feature enhancements are welcomed through pull requests.