Bringing Sports Action to Life in 3D: A New Way to Watch the Game

The Challenge of 2D Broadcasts

Sports broadcasts often present athletes in a flat, 2D manner, obscuring the true 3D motion of their limbs. Experts traditionally use field lines to determine camera positions and scene geometry, but zoomed-in shots remove these crucial markers, complicating accurate 3D data collection.

Introducing Partial Sports Field Registration

A groundbreaking method called partial sports field registration addresses this issue. By identifying a set of compatible camera positions and optimizing both the 3D pose of athletes and camera placement, researchers can now track running movements on a 400m track in 3D. This innovation marks a significant leap in sports analysis.

Testing the Method

To validate this approach, a synthetic dataset of over 10,000 images was created. It includes diverse viewpoints, running styles, and body types, exposing the limitations of current 3D pose estimation methods. The goal is to refine these methods for broader application in sports broadcasting.

Combining 2D and 3D Advances

This method integrates advancements in 2D human pose estimation and camera calibration, paving the way for large-scale kinematic datasets. These datasets promise to enhance sports analysis and broadcasting.

Open Access for Research

The synthetic data and code used in this research are publicly available, encouraging further collaboration among researchers and developers to improve 3D pose estimation techniques.