Soccer Player Identification in Live Streams

• R&D for a novel deep-learning solution to the problem of real-time identification of soccer players in

live/broadcast videos.

• Created an In-house Soccer players dataset extracted from Premier League and LaLiga videos.

• Developed a real-time player tracking pipeline in GPU-accelerated PyTorch using DeepSort with

YOLOv5 backend, reaching over 85% accuracy.

• Implemented Jersey number detection using YOLOv5, fine-tuned on our Soccer Player Dataset,

achieving over 90% accuracy.

• Implemented Jersey number recognition using ResNet trained on SVHN and fine-tuned on our

Dataset, achieving 87% accuracy.