Single-pixel imaging of variable-speed rotating objects via geometric moment centroid localization

Abstract

In single-pixel imaging of rotating targets, uncertainty in the target's angular position frequently causes misalignment between the projected illumination patterns and the object's spatial distribution, thereby degrading image quality, or even making imaging impossible. We propose a single-pixel imaging method for variable-speed rotating objects based on geometric moment centroid localization. The instantaneous rotation angle is estimated directly from the intensity sequence, eliminating the need for external sensors or synchronization hardware. A cyclic projection scheme—comprising three localization patterns followed by one Hadamard pattern—enables angle estimation through first-order geometric moments, allowing consistent-angle Hadamard measurements to be selectively retained for reconstruction. Experimental results demonstrate stable, high-quality imaging across rotation speeds as high as 6000 rpm, even under significant speed variations. At a 2 kHz projection rate, the method achieves an SSIM of 0.8402 with only 30% sampling, underscoring its robustness to undersampling and its practical viability in dynamic imaging applications.