Using the Discrete Fourier Transform for Character Motion Blending and Manipulation - a Streamlined Approach

Motion capture data allows natural-looking motion to be bestowed upon simulated characters. Research has sought ways of extending the range of motions it can reproduce. One such method involves blending between captured sequences in the frequency domain. This paper streamlines the approach taken by similar previous work. Higher efficiency is obtained both by shifting computations from runtime to pre-processing and by using a simpler technique, which is also more flexible allowing the method to be used for a greater range of motions. Furthermore, the already-known use of a triangular network defining a continuous blending space is instead presented as an adjustable interface element which is both intuitive and more flexible than applied to earlier work. As before input data may be sparse yet still allows the creation of a continuous spectrum of subtly varying motions, enabling characters to integrate well in their environment. Weighting calculation, blending and Fourier synthesis of realistic-looking motion using five harmonics requires 0.39 µs per degree of freedom for each frame in the created sequence - a one-off cost incurred only when blending ratios change. This figure can be improved further using the proposed level-of-detail adjustments, which, combined with its small memory footprint, makes the method particularly suitable for the simulation of crowds.