| 摘要 |
A method for range scanning consists of projecting a sequence of radiation patterns onto a scene, capturing images of the scene, determining correspondences between image features and projection pattern features, and computing a range image of the scene by triangulation of the correspondences. Novel projection patterns allow tracking of pattern features between images, so that range images can be computed for moving scenes in real time. Global assumptions about surface continuity or reflectivity of the scene are not required for successful range scanning. Projection patterns may be based on two-stripe codes, parallel stripes of bright and dark intensity. In two-stripe coded patterns, pattern features are the boundaries between stripes, which take on values of either on-on, on-off, off-on, or off-off. A particular boundary's value over the entire pattern sequence defines its code, used to determine correspondences between the boundary image and its location in the projection pattern. The method is also useful for obtaining three-dimensional models of static objects. The object is slowly rotated while being imaged, and all resulting range images are aligned and merged to form the three-dimensional model. A system implementing the method can be mounted on a robot or vehicle for navigation in unknown environments.
|
