Animated virtual humans may rely on full-body tracking system to reproduce user motions. In this paper, we reduce tracking to the upper-body and reconstruct the lower body to follow autonomously its upper counterpart. Doing so reduces the number of sensors required, making the application of virtual humans simpler and cheaper. It also enable deployment in cluttered scenes where the lower body is often hidden. The contribution here is the inversion of the well-known capture problem for bipedal walking. It determines footsteps rather than center-of-mass motions and yet can be solved with an off-the-shelf capture problem solver. The quality of our method is assessed in real-time tracking experiments on a wide variety of movements.

We describe in this paper a user study in the context of animated virtual human to compare the user selfperception of well known semi-autonomous Avatar and fullbody tracked Avatar. We aim at highlighting the advantages and limitations of those methods during various walking phases. The participants walked inside 4 simulated environments with different obstacles. Those results are quantified through a virtual reality sickness questionnaire and a new questionnaire specialized on user perception and enjoyment of his avatar lower part. This study shows positive results for the semi-autonomous Avatar especially in a cluttered environment. The users maintain the same efficiency with both methods, they have no sickness issues after more than 45 minutes inside the simulation and they present better enjoyment results for environment with complex obstacles.