A Survey of Selected Machine Learning Methods for the Segmentation of Raw Motion Capture Data into Functional Body Mesh

The paper describes the progress in the research for the automatic inferring method of the body structure—functional body mesh. In the paper we investigate four motion measures and machine learning methods—variants of Gaussian mixture models, DBScan and Neural Networks. The results were analyzed both—quantitatively and qualitatively using complete and incomplete data, of healthy and impaired persons. All the learning methods were on par with the others, however, we identified cases for which certain method works better.