CAMERA Speaker Series welcomes Dr Marion Mundt
Expert Speaker Series
We are pleased to invite you to our next talk in the CAMERA speaker series. Dr Marion Mundt will speaking to us from University of Western Australia in Perth.
30th March 2023- 10am
As Marion is based in Australia we will be hosting this talk online via Zoom. If you wish to attend please email [email protected] for log in details.
On-field motion analysis: repurposing motion capture datasets and training machine learning models to bring the lab to the field
Since the emergence of biomechanics as a modern scientific discipline, multi-scale analysis of human motion has been locked to laboratory environments. Predominantly this has comprised multiple near infra-red cameras and retro-reflective spherical markers affixed to an athlete’s body to record 3D motion kinematics, collected concurrently with ground embedded force plates recording ground reaction forces (GRFs) for the calculation of joint kinetics via inverse-dynamics methods. Consequently, there exist large 3D motion capture databases of commonly performed movements like running or sidestepping, and other unique smaller 3D datasets on specialised populations of complex movements and use cases.
In the ongoing efforts to enhance performance and prevent injury, 3D analysis in laboratory settings serves as the analysis approach of choice owing to greater accuracy, reliability, scale, and quantum, of the higher resolution data collected in that environment. Yet the laboratory setting does not easily allow for the assessment of internal and external factors that contribute to athletic performance during on-field training and competition environments. This data quality versus ecological validity trade-off has escalated the call to shift sports biomechanics data collection practices from the lab to the field. To facilitate this shift, wearable sensors and video-based motion analysis supported by machine learning algorithms have become increasing popular tools.
This presentation will provide insight into current machine learning applications in motion analysis and present methods to repurpose historical motion capture datasets to support both, wearable sensor and video-based analyses.