This paper was written as a seminar work for the lecture 193.179 Seminar in Visual Computing at TU Wien 2025W.
The final version of the paper can be downloaded here
As well as the slides for the presentation
3D Motion capture for the professional industry is traditionally being done with a marker based approach in complex studio setup including multiple cameras and special suits to achieve high accuracy. While this might suit certain needs such as film production and offline computer animation, other use-cases require real-time processing and high mobility instead of high accuracy. Furthermore, expensive investments in a motion capture studio create a high entry barrier for smaller organizations and individual creators. Therefore, this seminar papers tries to find a meaningful balance between high-cost high accuracy setups and low cost high mobility setups.
Human motion capture covers a large research field, various kinds of input data, output data a lack of common terminology and standards make it hard to compare approaches. To give an overview of the possibilities in motion capture methods 5 different methods are examined in detail. Benchmark datasets are presented alongside different output file formats and evaluation metrics. Afterwards, a use-case for 3D Live concerts using virtual avatars (VTuber) is described, and motion capture tasks are formulated. With a certain use-case and various kinds of approaches in mind a qualitative and quantitative comparison of 8 different methods is done. Lastly, a potentially suitable motion capture method is described, and current challenges are explained.