Qualitative comparison between marker-based and video-based human pose estimation in the context of imitation learning

The Master's thesis explores whether video-based 3D human pose estimation can serve as an effective alternative to marker-based motion capture for imitation learning from human demonstration. 
The data will be recorded through a study using multiple modalities, it will be conducted with a custom imitation learning scenario and evaluated based on pose estimation quality, action separability, action recognizability, and trajectory mapping quality. 
The study will assess the data quality in the context of imitation learning, aiming to determine the practicality of video-based techniques for dynamic robot learning.