Dual-Arm Motion-Compensated Hyperspectral Single-Pixel Imaging

Single-pixel imaging offers a cost-effective strategy for high-resolution imaging over a wide range of electromagnetic frequencies, but suffers from motion artifacts due to its sequential acquisition process. In this work, we propose a new framework for dynamic single-pixel imaging that is particularly suited for hyperspectral imaging. First, we introduce a hybrid dual-arm device combining a hyperspectral single-pixel camera and a conventional imaging arm, which allows accurate motion estimation during acquisition. We then reformulate the reconstruction problem by compensating for motion, thereby reducing the dynamic problem to a static reconstruction task over an extended field of view. Two different discretizations -warping either the illumination patterns or the image to be reconstructed- are proposed, along with an in-depth analysis of their trade-offs. Through extensive numerical simulations and real-world experiments, we demonstrate that warping the image rather than the patterns leads to superior reconstruction quality. In addition, extending the field of view beyond that of the single-pixel camera significantly mitigates model mismatch and improves image fidelity. The proposed method achieves low computational cost while maintaining theoretical rigor, providing a practical and robust solution for dynamic single-pixel imaging. Open-source implementations are provided to facilitate reproducibility and future research.