Automating the collection of spatio-temporal and kinetic performance parameters using computer vision

Synopsis

The collection of spatio-temporal and kinematic performance data is currently a manual, laborious and timeconsuming post hoc process that relies on biomechanics domain knowledge to define the correct frames to analyse, annotate joint centres, and calculate relevant performance parameters.

Dr Marion Mundt was successful in an AIS grant application that seeks to develop an automated computer vision-based assessment tool, drawing on advanced machine learning techniques, to provide spatio-temporal and kinematic performance parameters of the high jump approach from 2D video.

To achieve this, the detection of gait events from 2D videos and the accuracy of pose estimation keypoints need to be assessed and the influence of those on performance parameters established. Using this information, a biomechanically-informed pose estimation model can be developed to provide relevant performance parameters required by the athlete, coach, and performance support staff.

Biography

Dr Marion Mundt is a Research Fellow in the UWA Tech & Policy Lab at The University of Western Australia, working with the Australian Institute of Sport to use and validate machine learning techniques to estimate kinematic and kinetic motion parameters from standard two-dimensional video. She received her PhD in Sport Science from the German Sport University Cologne for the application of artificial intelligence to motion analysis using inertial sensors.

In 2022, she received the Hans Gros Emerging Researcher Award from the International Society of Biomechanics in Sport for her work in ‘bridging the lab-to-field gap