Scene-level passive polarization 3D imaging

Scene-level passive 3D imaging under natural conditions is highly challenging yet urgently demanded. Polarization 3D provides possibility but impeded by two major obstacles brought by natural large scenes: discontinuities of multiple targets and dynamic reconstruction. This study proposed a scene-level passive polarization 3D imaging method, integrating binocular stereo and polarization. We abstract the discontinuous targets reconstruction into a minimization problem. The pixel-level normal direction from polarization and the absolute scale information from binocular stereo vision then work as mutual constraints for iterative optimization of the problem. By iterating for final solution, the challenge of reconstructing discontinuous targets was tackled, and true depth was also recovered. The true depth then provides a reference for solving inter-frame scale inconsistencies which hinders dynamic reconstruction by the designed scale normalization strategy which globally aligns multi-view measurement data. Scene-level 3D structure was finally reconstructed through multi-frame point cloud fusion. We showcase wide-scene, high-accuracy passive video reconstructions on natural field scenes. Our passive polarization stereo represents a major advancement in scene-level 3D imaging and may find broad applications in fields requiring passive 3D imaging solutions.