Can Markerless Tracking of Human Motion Revolutionize the Analysis of In-game Sports

Performance?

W Scott Selbie
Theia Markerless Inc; C-Motion Inc; Queen’s University (Canada)

Contact: kate.cowperthwaite for Zoom Coordinates

I will present the state of the art 3D tracking of human motion suitable for biomechanical analyses based on
using Deep Similarity Learning to identify uniquely all players in the field of view of the cameras, Deep
Neural Networks to identify the location of anatomical features of each player in the multiple video images,
and multibody optimization to consolidate this data into a mathematically observable estimate of the 3D
position and orientation (pose) of a personalized model of each player. Such systems do not require
assumptions of symmetry, behavioural rules or physical constraints as they directly estimate the 3D pose. The
nature of the technology lends itself to the vast amounts of accurate and meaningfully consolidated data
required by the burgeoning field of sports analytics. I will present experimental results of the critical issues of
accuracy, repeatability and reliability of the pose estimation in a controlled laboratory setting and speculate on
these same issues in live game recording.