Estimating surface shape and extending known structure using specular reflections

In this paper a method for the shape estimation and structure extension of a surface using information from specularities is proposed. The structure of the scene is obtained from an image sequence using standard structure and motion techniques. If there are specularities in the scene, there may be parts where the scene geometry cannot be determined. The method proposed in this paper extends the known structure over these regions. Methods for determining the light source position are introduced and together with the reflections, this gives constraints on the surface normals. These constraints are complemented by a smoothness condition in order to estimate the surface shape unambiguously. The viability of the proposed method is demonstrated in experiments with the real data from image sequences.