AI tool for the
indoor sports

See our innovative product

ISVP.AI R&D stages of preparation

Data preparation

Training dataset based on football matches videos:

  • Marking ball goals, referee or referees, players on each frame of the match recorded in real condition 

Match recordings preprocessing module: 

  • Dividing video into single frames
  • Detection of field, filtering out irrelevant fragments

System integrations

Target architecture of ISVP.Al system:

  • Based on cloud services (Amazon, Google or Microsoft or suppliers focused on solutions using machine learning)

Training of the selected algorithms:

  • Training with the entire training dataset based on the target architecture

Neural network training

Image instance segmentation module:

  • Detecting goal gates, players and ball on each frame of the video + based on convolutional neural networks

Initial training of the chosen neural network architectures:

  • Training only on a subset of the total training dataset 
  • Calculations performed on the GPU 

Detected people objects classification module:

  • Assigning classes of the players and other objects . classifying players to the teams

Web application ans optimization

Preparation of a web application:

  • Possibility of uploading the match recording by the target recipient 
  • Analysis using the ISVP.Al engine
  • Presentation of a produced video

Optimizations in the source code:

  • Improving system performance and efficiency by adjusting hyperparameters of the system or rewriting computationally demanding pieces of code in C language

Tracking and analysis

Movement tracking:

  • Tracking the location of objects from the first frame
  • Redetection of the objects (by neural network) In case the video tracking mechanism fails

Detection of football match events:

  • The beginning and the end of the match, the first and second half and unplanned breaks

Tools for sport
industry experts
that think differently

Join our contellation and book a demo