Automatic Key-Frames Extraction of Humanoid Motions

3D authoring tools provide a wide variety of motion editing functionsto users to make motions appear more plausible and natural. Animatorsobviously need to spend much time and invest considerable efforts to editgreat amounts of frames for the sake of obtaining their desired animationresults. Therefore, animators try to reduce their burdens by picking up theappropriate key-frames, editing these chosen key-frames, and interpolatingthem for their desired results. Even animators with sufficient experiencein animation may need to use several number of times to gain the appropriatekey-frames. In this paper, we present a method to overcome aforementionedkey-frame problem by automatically picking up the most significant keyframesfrom a continuous motion clip. Users only need to input a targetmotion clip in Biovision Hierarchy (BVH) format and provide a series ofparameters, and the proposed method will extract the most significant keyframesfrom target motion clips automatically. In order to verify the qualityof extracted key-frames, we directly interpolate the extracted key-framesand compare them with the input motion clip. The experimental results suggestthat the proposed method is less labor-exhaustive, and it decreases theoperation time without sacrificing original motion features.