One Shot 3D Reconstruction by Observing Multiple Spherical Mirrors

Since 3D information is helpful to 3D scene observation and recognition, 3D information acquisition and display technologies are used in various fields. Several methods to acquire 3D information such as laser scanner (LiDAR), time of flight (ToF) sensor and 3D photogrammetry from multi-view images are under active development. However, the common disadvantage of these methods is that they require a special equipment or large-scale observation system. This paper proposes a method to estimate 3D information from a monocular image that pseudo captures multi-view images. It utilizes a catadioptric imaging system given by multiple (at least two) curved mirrors and stereo-vision technique. The catadioptric imaging system has a wider field of view than that of a typical camera based on perspective projection. It is suitable for reconstructing 3D information of open spaces with a monocular image. We confirm the quality of estimated 3D information by generating arbitrary viewpoint images.