A method of detecting 3D object orientation by multiview models

Abstract

The problem of determining the approximate orientation of an object from multiview models is considered for the case in which the number of models if fixed but the viewing angles are arbitrary. The multiview models include the height and width of each projected image of a 3D object, topological information based on the edge image, some region information and the angles of each projected image. Assuming there is only one object being considered, an image is input in an arbitrary viewing direction (camera focal length and the distance between the camera and the object are known) and processed to obtain the edge image. The edge image coordinate data are transformed into normalized data with respect to focal length and distance. The height and width of the object are compared with the models' height and width to determine a set of candidates, and topological and region information are used for further matching to detect the approximate orientation. In their experiment the authors built the multiview models from 16 viewing angles, each roughly equal to 22.5 degrees .<>