An evaluation of a cost metric for selecting transitions between motion segments

Designing a rich repertoire of behaviors for virtual humans is an important problem for virtual environments and computer games. One approach to designing such a repertoire is to collect motion capture data and pre-process it to form a structure that can be walked in various orders to re-sequence the data in new ways. In such an approach identifying the location of good transition points in the motion stream is critical. In this paper, we evaluate the cost function described by Lee et al.15 for determining such transition points. Lee et al. proposed an original set of weights for their metric. We compute a set of optimal weights for the cost function using a constrained least-squares technique. The weights are then evaluated in two ways: first, through a cross-validation study and second, through a medium-scale user study. The cross-validation shows that the optimized weights are robust and work for a wide variety of behaviors. The user study demonstrates that the optimized weights select more appealing transition points than the original weights.