Fast Computation of Transition Points for Motion Graph

Motion capture data analysis and reuse is an active topic in computer graphics research. Motion graph is a widely used data structure to organize raw motion capture data. A basic preprocessing step before building a motion graph is to compute pose similarities between all pairs of frames in motion data and find out local minima as candidate transition points. However, this task can be very time consuming. In this paper, we introduce a simple strategy to compute local minima efficiently. Instead of fully computing all pairs of frames in motion data to compose the resulting motion distance map, our method uniformly sample the distance map by skip frames according to an user defined sampling interval. The local minima in sampled distance map are thus restricted on sampling grids, and we introduce error metrics to evaluate the loss in accuracy. We use this error metrics as guidance to adjusting the positions of candidate transition points. We also provide an simple method to determine appropriate sampling interval. Result shows that our method can compute transition points very fast and get good approximation of transition points without our methods.