Structured-light system full calibration from an unknown box

Calibration of structured-light (SL) system is a prerequisite step for 3D reconstruction. Previous calibration methods often rely on observing a planar pattern with known geometry from multiple views, which is time-consuming and cumbersome. Or some methods follow a self-calibration pipeline with an unknown object, but too many a priori assumptions, such as a known principal point for a camera or/and projector, make these methods only a “partial” calibration. To this end, we propose a full calibration method with an unknown box, where, in addition, only a single view observation suffices, i.e. the box needs not to be observed in different views. The proposed method consists of two stages. In the first state, we firstly derive the imaging geometry of a box or cuboid, based on which both camera intrinsics and parameters of the cuboid are recovered. In the second stage, the projector parameters are then estimated from the recovered cuboid and the accurate correspondence between the camera and projector views. Furthermore, since both stages solve linear problems, the proposed method is also time-efficient. Extensive experiments were conducted, of which the results demonstrate the effectiveness of the proposed method, which will find wide applications in daily-life use, particularly those more emphasizing on the flexibility, easy-to-use and low cost, such as AR/VR, 3D reconstruction in entertainment use.