Shape from polarization via ultra-low-sampling single-pixel imaging with decoupled spatial and polarimetric dimensions

Shape from Polarization (SfP), a passive three-dimensional (3D) imaging method with high precision, long detection range, simplicity, and low cost, holds significant application value in surface defect inspection, virtual reality, and robotic vision. However, traditional SfP techniques face high storage demands, large bandwidth transmission, and massive data processing burdens, leading to high computational load, energy consumption, and cost, which severely limit their mobile deployment. To address this, we innovatively propose a single-pixel ultra-low sampling SfP system based on decoupled spatial and polarimetric dimensions, combined with a deep fusion network. It acquires raw target data at an ultra-low sampling rate (0.39%) via deep undersampling. Experimental results demonstrate that, at a pixel resolution of 1024 × 1024 , while maintaining reconstruction accuracy comparable to conventional methods, the raw per-frame data volume is reduced from 4.00 MiB of traditional full-sampling with four polarization states to only 0.0156 MiB. Correspondingly, the storage occupation and transmission bandwidth are reduced by 256 times, greatly alleviating the pressure of data storage and transmission at the acquisition terminal. This low-data-volume, low-cost approach effectively addresses the technical bottlenecks of rapid image data transmission and processing in bandwidth-constrained imaging scenarios, thereby offering a new technological pathway for 3D imaging on resource-limited platforms in static environments.