AbstractWe have used the intersection of polarised reflectance contours to determine surface orientation. This information is used to both to recognise and to determine the orientation of dark, specular, dielectric objects. The system uses three images of an object viewed through a plane polarising filter at different rotations.
The objects are described by their Extended Gaussian Images (EGIs). The process always provides a description of only that part of the object which is visible to the camera, i.e. a viewer-centred description. This description of the object takes the form of a histogram. The cells of the histogram represent surface orientations, and the frequency in each cell is the visible area of the object at that orientation.
This sensed histogram is compared with a database of pre-computed, viewer-centred, model orientation histograms. The best match determines the type of object and its orientation in space. Matching with a pre-computed database is fast, efficient and determines orientation to within an average uncertainty of less than 7 degrees of arc from fewer than 30 comparisons.
|Date of Award