3D volumetric reconstruction with a catadioptric stereovision sensor

A stereoscopic catadioptric sensor and associated algorithms are presented to provide a relevant solution to the problem of 3D reconstruction of unknown environments. A novel calibration methodology is detailed, establishing a discrete relation between 3D points and the corresponding pixels. The 3D reconstruction adopted is an adaptation of the classic volumetric scene reconstruction technique. Complete voxels are projected onto the image planes and consistency measurements are computed on pixel surfaces. To ensure fast processings, 3D/2D matchings are processed offline and results are stored in Look-Up Tables (LUT). Calibration results are given and 3D reconstruction algorithms are validated on synthetic images. Some results obtained for real scenes are provided.