Davies, C.J., Nixon, M.S. and Carter, J.N.
A New Hough Transform Mapping for 3D Imaging via Coloured Spots.
Full text not available from this repository.
Three dimensional (3D) mouth shape extraction is vital to enhance models of speech production, since lip movement is important in speech production and perception. Many techniques have been developed for 3D surface shape sensing, but are restricted to static or slowly moving objects. Video-rate triangulation based structured light systems enable 3D surface data acquisition for dynamic scenes. The surface depth information is calculated from a single image of the scene illuminated by a single projected pattern. Colour encoded stripes and encoded square binary pattern elements have previously formed the basis of the projected patterns. In a new structured light system, the projected pattern is an array of colour encoded circular spots. These spots are deformed by the underlying surface, both in position and shape, and are perceived as ellipses. This enables both the surface position and the surface orientation to be captured simultaneously from each imaged spot. Using smooth-shape pattern elements avoids the inherent potential error of other systems in the presence of surface step edges. Feature extraction is made more efficient by a-priori knowledge. In the new 3D sensing system the centre position of each imaged spot is constrained to lie on an epipolar line, constrained by the working volume of the system. The imaged shape of each spot depends on the orientation of the illuminated surface. These constraints enable the formulation of a Hough Transform (HT) to extract directly the position and orientation of the surface in the scene via the imaged spots. The HT has an accumulator space of only three parameters, these describe the 3D surface's position and orientation. Results verify the new formulation of the HT and demonstrate how this new structured light system can be used to recover 3D surface shape.
Actions (login required)