Title: Deriving Depth via Dual Differential Defocus and Stereo Consensus

Abstract: This paper presents D^3S Consensus, a closed-form, physics-driven algorithm that integrates depth-from-defocus (DfD) with stereo vision to deliver precise depth estimation across an extended operational range, surpassing the limitations of traditional camera depth-of-field (DoF). By processing a pair of stereo images captured with dual defocus, the system leverages a new theoretical framework known as Dual Differential Defocus (D^3) to calculate an overdetermined set of depth values. These are combined with stereo data in a coupled manner. To ensure reliability, the method selects the most robust depth prediction by enforcing consensus between these two physically distinct cues, effectively filtering out inaccurate estimates. Our analysis indicates that D^3S matches the working range of existing triangulation-based systems under identical error tolerances, but requires a baseline that is ten times smaller. This efficiency allows for the development of compact, passive binocular rangefinders that are significantly more miniaturized than conventional stereo or DfD setups. We showcase the first D^3S prototype, which features an EFL of 12 mm and a baseline of just 4 mm. Capable of snapshot acquisition, it produces depth maps with a resolution of up to 900 x 1800 pixels, maintaining a mean absolute error of 1 cm within a range of 0.3 to 1.64 meters. These results demonstrate superior accuracy compared to certain commercial stereo cameras that possess considerably larger physical dimensions.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC