Dual Quaternion Discrete Linear Canonical Transforms: Theory, Fast Algorithms, and Chirp-Adaptive Filtering for Rigid-Body Pose Signals
Abstract
We develop one- and two-dimensional dual quaternion discrete linear canonical transforms (DQ-DLCTs) for rigid-body pose signals, with left-, right-, and two-sided formulations that account for noncommutative kernel ordering.
We establish inverse relations, kernel unitarity, energy preservation, perturbation stability, and spatial and frequency convolution relations.
A chirp--Fourier--chirp factorization enables fast forward and inverse algorithms using complex FFTs, with two-dimensional complexity $\mathcal O(N_1N_2(\log N_1+\log N_2))$.
For pose denoising, we introduce a two-stage chirp-adaptive method that estimates rotation before refining translation while enforcing unit-dual-quaternion constraints.
Numerical experiments demonstrate agreement with direct transforms, substantial computational gains, and lower pose errors than the tested baselines on measured sequences and constructed two-dimensional fields.
A downstream $\mathrm{SE}(3)$ synchronization study further demonstrates the potential of the proposed preprocessing approach, with performance depending on graph configurations.
These results support efficient and geometry-aware processing of rigid-body pose signals.