Scene-Level Passive Polarization 3D Imaging
Scene-level passive 3D imaging under natural conditions remains a critical unmet need, as established techniques such as structured light, LiDAR, and active stereo rely on controlled illumination and scanning, limiting their applicability to large, dynamic outdoor environments. Passive polarization 3D imaging offers inherent advantages for long-range, high-precision reconstruction but is fundamentally impeded by two obstacles: the π ambiguity of the azimuth component of surface normals and the discontinuity of multiple targets within a scene. This study introduces a scene-level passive polarization 3D imaging method that integrates binocular stereo vision with polarization cues. The reconstruction of discontinuous targets is formulated as a minimization problem, where pixel-level normal directions from polarization and absolute scale information from binocular stereo serve as mutual constraints for iterative optimization. This framework resolves the discontinuity challenge and recovers true depth. A scale normalization strategy globally aligns multi-view measurement data, eliminating inter-frame scale inconsistencies that hinder dynamic reconstruction. Multi-frame point cloud fusion yields the final scene-level 3D structure. Experimental validation on natural field scenes demonstrates robust, wide-scene, high-accuracy passive video reconstructions with centimeter-level precision. This passive polarization stereo approach represents a significant advancement in scene-level 3D imaging, with potential applications in autonomous navigation, environmental monitoring, and cultural heritage documentation.