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

Area of Science:

• Computer Vision
• Robotics
• 3D Sensing

Background:

• Structured light systems using camera-projector pairs (CPPs) excel at 3D reconstruction, especially in textureless indoor environments.
• Existing methods require pre-calibrated intrinsics or known reference objects, limiting practical indoor applications.
• Recovering CPP intrinsics from minimal views without known objects is a significant challenge.

Purpose of the Study:

• To develop a reliable calibration framework for unknown CPPs enabling direct 3D reconstruction.
• To overcome limitations of existing methods by eliminating the need for pre-calibration or known objects.
• To achieve robust and accurate intrinsic recovery using readily available indoor structures.

Main Methods:

• Exploiting an unknown cuboid corner (e.g., a room corner) to derive sufficient geometric constraints from just two views.
• Utilizing the known camera principal point to simplify multi-parameter estimation into a univariable optimization problem.
• Developing a calibration framework for direct 3D reconstruction with an unknown CPP.

Main Results:

• Achieved stable and accurate recovery of camera-projector pair intrinsics.
• Demonstrated superior robustness and reconstruction accuracy compared to traditional and learning-based methods.
• Validated the framework's effectiveness through extensive experiments.

Conclusions:

• The proposed method offers a simple, reliable solution for intrinsic recovery and 3D reconstruction with unknown CPPs.
• The framework's ability to extend to passive settings shows potential for sparse-view structure-from-motion.
• This approach enhances the practicality of 3D reconstruction in diverse indoor scenarios.