High-Speed Analytics: Tracking Table Tennis Movements with RF-DETR and OC-SORT

Developing a computer vision solution for table tennis involves creating a high-resolution and spatially accurate system to track the fast-moving ball, which can exceed speeds of 100 km/h, and generate precise movement data for analysts and coaches. The process, traditionally reliant on manual logging and video review, is now enhanced by automated vision pipelines using the Roboflow Platform. The system is built by developing a prototype tracking system that converts match footage into structured metadata, starting with setting up a development environment and sourcing a suitable dataset. Through meticulous steps such as labeling, annotation, and training the RF-DETR model, which is optimized for low-latency inference, the system ensures quick processing necessary for real-time tracking. Data partitioning, preprocessing, and augmentation further refine the model's accuracy, ultimately achieving high precision and recall metrics. By employing the OC-SORT tracking algorithm, the system maintains consistent tracking even when the ball is briefly obscured, and a series of visualization blocks transform raw data into a broadcast-ready overlay. For seamless deployment, users can manually configure the system or use the Roboflow Agent to automate the workflow, offering a sophisticated analytics framework that logs ball speed, spin trajectories, and player positioning with precision.