Analysis of player tracking data extracted from football match feed

Abstract

Data analytics and AI have become extremely relevant in today’s football landscape. Data is benefiting clubs in gaining a competitive advantage on and off the field by empowering them to harvest information for improving player performance, decreasing injuries, and increasing commercial efficiency. Tracking data in football, that is the data of x, y coordinates of all 22 players on the pitch every second adds a lot of value in terms of a team’s decision-making and recruitment strategy. Via the use of this data, each player’s decision-making ability is also measured. The players and the ball in each frame are identified using YOLOv5, which returns their coordinates. These detections are then passed to DeepSORT, which assigns IDs to each player and keeps track of the frame by frame, by feeding each player detection to the model. The K-Means model is employed to determine the jersey color of the players to identify the two teams. Finally, the detected coordinates are multiplied with the homogeneous matrix computed using the Sports Camera Calibration through the Synthetic Data paper approach to accomplish the perspective transform. The mathematical model hypothesized and implemented uses pitch control and expected threat to assess each player’s decision-making ability, which will be a strive to enhance the recruiting of players.

Keywords

• computer vision
• perspective transformation
• yolov5
• deepsort
• object localisation
• object tracking
• data analytics
• pitch control
• expected threat
• voronoi diagram