Indoor 3D reconstruction using an unknown camera-projector pair.

Structured light systems based on a camera-projector pair (CPP) are widely used for indoor 3D reconstruction, particularly in textureless environments where passive methods often fail. However, existing approaches typically rely on pre-calibrated intrinsics or multi-view self-calibration with known reference objects, which limits their applicability in practical indoor scenarios. Recovering CPP intrinsics from only two views without any known objects remains a challenging problem. In this paper, we present a simple yet reliable calibration framework that enables direct 3D reconstruction with an unknown CPP. By exploiting a commonly available indoor structure-an unknown cuboid corner (C2) (e.g., a room corner), we show that sufficient geometric constraints can be obtained from only two views. With only the camera principal point known, the initially coupled multi-parameter estimation is reduced to a univariable optimization problem, resulting in stable and accurate intrinsic recovery. Extensive experiments demonstrate that our method consistently outperforms both traditional and learning-based alternatives in terms of robustness and reconstruction accuracy. Moreover, the framework can be naturally extended to passive settings without active illumination, showing promising potential for sparse-view structure-from-motion applications.