Tracking Athletes in video
The business goal
Our client is a telecom company from South Korea. The company records and analyses video of sports events and in order to improve their analytics. They decided to use Computer Vision. This very project was done as a part of the preparation for the Olympic Games in Seoul.
The goal of the project was to analyze video of athletes using multiple cameras and identify/detect the objects in real time, while the analysis system detects and tracks athletes appearing on video in real time. Detection of athletes consists of finding people on video that have a predefined look and wear a specific type of clothes.
We used Python, OpenCV, Faster RCNN and TensorFlow as a Computer Vision/algorithm part. Specifically, the deep learning model was based on TensorFlow Framework; as a tracking system we used custom algorithms based on Kalman filters. We also created a communication layer between the system and video server using high-performance custom protocol based on TCP-sockets usage.
As a result, our client was able to provide more informative analytics reports to its audience.