| A well-known technique for object ranging by computer vision requires a stereo pair of cameras and relies on simple triangulation. Each object point to be ranged must be imaged by both cameras. The orientations of the cameras must be known, as well as the correspondences between pairs of camera points. That is, it must be known that two camera points, one in each camera, are images of the same object point before they can be used to triangulate the object point. Obtaining this correspondence information is not a trivial task.; Another ranging technique involves replacing one of the stereo cameras with a projector which actively illuminates the scene with a light pattern. For example, holes (or grooves) in a template are projected as known beams (or sheets), incident on a surface to be ranged, and the illuminated points (or stripes) are imaged by the camera. The correspondences between the projector beams (or sheets) and the camera points (or stripes) are required for triangulation of the illuminated surface points (or stripes). Even in such a structured environment the required correspondence information is difficult to obtain.; This research describes how the pattern illumination technique for range mapping can be refined so that the correspondence information is obtained in a novel, elegant fashion. In fact, the locations of the projected beams need not be known beforehand as with other techniques. The only requirement is that the camera and/or the projector must move in a trajectory of known positions and orientations. One simple but robust motion which allows precise ranging of numerous object points is rotation of the projector about any axis oriented in the general direction of projection. |
