Sport-analytics

Sport analytics using AI.

Develop an AI solution for sport analytics (football), which will:

• Detect players
  • on single frame
  • splitted window frame
• Classify players per team
  • using machine learning techniques
• Detect player's number in jersey (not done)
  • used pre-trained yolo model in SVHN dataset
• Image mapping into 2D football court (not done)
  • Generate dataset with frame - homography map
    • Generate database of homography maps, by zooming, tilting, etc.
  • Train GAN network to generate edge map images given frame input
  • Using HOG descriptor, find the best matching homography maps of database to the generated one
  • Plot image in 2D football court.

Install

This project requires Python and the following Python libraries installed:

• NumPy
• Pandas
• matplotlib
• scikit-learn
• Tensorflow
• PyTorch

You will also need to have software installed to run and execute a Jupyter Notebook or run it in Google colab.

If you do not have Python installed yet, it is highly recommended to install the Anaconda distribution of Python, which already has some of the above packages and more included.

Pre-trained networks - used tools

For:

• Players detection used YOLO (You only look once) which is a real-time object detection system.
• Video mapping into 2D football court used pix2pix CAN

Datasets

For:

• Video mapping into 2D football court, used FIFA 2014 World Cup Dataset
• Detect player's number in jersey, used SVHN Dataset

Files

For:

• Main notebook check Football_analytics.
• Number's detection check try check Detect_numbers.
• video generation, check video_playback.py

References

[1] Sports Video Analysis using Machine Learning

[2] Automatic Birds Eye View Registration of Sports Videos

[3] YOLO digit detector

[4] Pix2Pix Pytorch

[5] Finding the dominant colors of an image