Recognition and re-synthesis of 3D human motion with personalized variations

This paper proposes a 3D view-invariant human action recognition method based on Hidden Markov Models. The natures of the actions, as well as the characteristics of the actors and different performance styles have been successfully recognized. The results have been compared to nearest neighbor and similarity search based recognition for further evaluation. Also the research addresses the problem of re-synthesis of motion. Transformation of moods, genders, and other characteristics of the actor have successfully been carried out, and the entire action has been re-synthesized for various purposes such as animation.