Estimation of shape, lighting, and reflectivity from stereo and shading information

This thesis is concerned with the combined use of stereo and shading information for the purpose of three-dimensional surface shape estimation. The applicability, resolution, and accuracy of each individual method is examined and compared with that attainable in a combined approach. Integration is examined from standard and stochastic perspectives. The standard approach involves use of stereo information as an initial estimate of the surface shape from which parameters of the shading method such as the light source direction can be estimated. Issues of lighting and reflectivity estimation are addressed and solutions are advanced. In particular, general methods for estimation of the light source distribution and reflectance map are proposed and tested. Colored lighting and color variant reflectance functions are examined as possible sources of additional shape information. A statistical approach to surface shape estimation called 'shape from appearance' is also advanced. In this approach, stereo information is used indirectly as a source of statistical information. It is shown that through a scale change the resulting statistical model can be used to achieve higher resolution than is achievable through the direct use of stereo information.U of I OnlyETDs are only available to UIUC Users without author permissio