ICA-based interpolation of human motion

Generating realistic human motions from sparse pose descriptions are important for computer animation and virtual human applications. In this paper, we propose an example-based motion interpolation technique using independent component analysis (ICA). We extract a few independent motion signals from multi-dimensional human movement. The result of the extracted motion signal corresponds to the perceptually meaningful motion features that characterize the complex human actions. We can obtain the plausible motion by interpolating arbitrary sparse pose in ICA space. The detail of the subtle motion can also be recovered by motion texture mapping. The proposed method is also useful for human motion conversion, motion blending, and converting symbolic descriptions to continuous motion.