Football Analysis

Overview

This project analyses football matches using advanced computer vision techniques, including:

• YOLO for object detection (players, ball, referees).
• ByteTrack for tracking movements.
• KMeans and SigLip for team classification.

Key Components

1. Input and Output Processing

• OpenCV is utilized for:
  • Extracting frames from video inputs.
  • Applying visualizations on the processed frames.
  • Saving the processed frames back into video format.

2. Object Detection and Tracking

• YOLOv11n is used for object detection.YOLO
• ByteTrack is utilized for tracking players, ball, and referees. ByteTrack
• The initial dataset used was provided by Bundesliga on Kaggle. It is now available on Roboflow.

3. Team Classification

Workflow:

1. Player Cropping:

   • Player crops are extracted from video frames using YOLO detection boxes.

   

2. Background and Torso Clustering:

   • KMeans is applied to separate the background from player torsos.

| | |

| Before Clustering | After Clustering Background and Torso |

3. Team Color Classification:
   • The colors of players' shirts are clustered to distinguish between two teams.

How to Run

1. Install dependencies:

   pip install -r requirements.txt
   

2. Download required models from the models/ directory.
3. Execute the script:

   python main.py -i {input_video_path} -o {output_path} -r {results_path}
   

4. Outputs:
   • Results are saved as parquet files.
   • Processed frames and videos are stored in the specified output path.

Upcoming Features

1. Radar View:
   • Using mplsoccer to create a radar view for tactical analysis.

| | |

2. Enhanced Statistics:
   • Calculating player metrics such as:
     • Movement speed.
     • Passing accuracy and types of passes.
   • Generating heatmaps and advanced visualizations for deeper insights.