Ball position tracking in the cloud with the PGA TOUR

The article discusses the development of a modular pipeline by the Amazon Generative AI Innovation Center (GAIIC) for the PGA TOUR to track the position of golf balls on the green using computer vision and machine learning (ML) techniques.

Specifically, the article covers:

• The data provided by the PGA TOUR, including 3 days of continuous video from 4K cameras positioned around the green and annotated scoring data.
• The overview of the training and inference pipelines, which involve video ingestion, preprocessing, labeling, model training, and deployment.
• The player detection module using a pre-trained CNN to find people in the field of view.
• The player classification and ball detection module using a fine-tuned YOLO v7 CNN to classify players putting and detect the initial ball position.
• The ball path tracking module using a pre-trained ResNet CNN for motion tracking.
• The pipeline output, showing labeled video with people, putter, and moving ball tracked.
• The performance evaluation, including player classification, ball detection, and ball path tracking accuracy.
• Conclusion highlighting the successful development of the modular pipeline for ball position tracking.