Football Analytics Collection

About the Client

Our client is a US company that builds various applications for non-professional athletes to save memory records and personal achievements.

The client operates services powered by innovative technologies to provide excellent user experience to sports fans.

Business Challenge

The client’s goal was to make coaching, player development and recruiting effortless. They strive to bring professional football analytics to non-professional football clubs.

To achieve this, the client offers an easy-to-install and use solution armed with two cameras that uses a cloud platform for analytics extraction.

Solution Overview

Our team adopted the Quantum video analytics platform for people detection and tracking on a football field with further statistics gathering.

Our solution combines video, GPS sensors, and artificial intelligence into a single intuitive application.

The solution has three main components:

  • The people tracking system
  • The ball tracking system
  • The pattern recognition system

Project Description

We used the Quantum video analytics platform as the solution’s core. The system consisted of three components:

People tracking. We adopted the people tracking model to meet specific environments and high accuracy requirements. For example, high-quality panoramic images were sliced to increase processing performance and get qualitative data about the player position.

Ball tracking.  Developing this component was a challenge because of the high speed and small size of the ball. Our R&D team found a new way to detect it with another model that was based on fast-moving object detection technology.

Pattern recognition. Football match statistics include about 20 events like passes, goals, outs, etc. Humans can recognize these patterns fairly easily, but they have to be translated into math language for the computer.

The Quantum video analytics platform we used as the basis allows describing patterns based on behavior, relative positions, and object interaction. Based on an interview with our product owner and following his explanations, models were trained to recognize patterns like a human.

Let's discuss your idea!

Technological Details

The pipeline processes two 2,7K video streams with 20FPS and 80%+ accuracy on NVidia GTX 1080 Ti.

Complemented with personal GPS trackers, it demonstrates a stable accuracy of nearly 100%.

Python
Python
OpenCV
OpenCV
Tensorflow
Tensorflow
NVIDIA GEFORCE GTX
NVIDIA GEFORCE GTX

Connect with our experts