ankitsinghh12 / YOLOv5_Football_analytics

A state-of-the-art object detection system, built on the YOLOv5 framework, is employed to individually recognize and monitor players, the ball, sideline referees, and the goalkeeper in real-time using a television broadcast camera feed. The Bytetrack technology is integrated to augment and refine the model's performance! ⚽

Football Players Tracking on a pitch using YOLOv5 + ByteTrack

A YOLOv5-based object detection model that identifies and tracks the players, the ball, the sideline referees and the goalkeeper separately when provided with a TV broadcast camera feed. Bytetrack is further used to enhance the model's capabilities! ⚽

About

In my project, I utilized Roboflow Universe, an open-source computer vision dataset repository with over 100,000 datasets. We make use of two pretrained models one after another and apply custom annotations on the latter to get our output.

Useful links:

• YOLOv5 repository 🔗
• ByteTrack repository 🔗
• Roboflow Notebooks 🔗

Steps Involved:

• Setup🛠️
  
• Download data🛠️
  
• Install YOLOv5🛠️
  
• Install ByteTrack and other libs🛠️
  
• Custom annotator🛠️
  
• Detect ball possession🛠️
  
• Full video tracking🛠️
  
• Put everything together🛠️
  

Output!⚽

Sample Video Frame:

Output Video after first petrained model:

Output video after v2 pretrained model:

Frame after applying Bytetrack: