Application of Quaternion Scale Space Approach for Motion Processing

Scale space approach turned out to be a powerful framework for image processing purposes. In this paper we introduce proposal of a scale space concept generalized for datasets of quaternions e.g trajectories representing motion capture data. We introduce equivalents of non-linear and anisotropic simplification operators which exhibit a completely new set of properties. Three groups of applications are considered: trajectory smoothing, meaningful features detection and parameterized signal synthesis. We present a theoretical analysis and the results of numerical experiments.