3D reconstruction of objects in the presence of strong inter reflection or subsurface scattering using structured light.

Abstract

In this thesis a new structured light method is proposed. It is a line shifting algorithm in combination with error correction based on scene continuity. The method initially gathers a large number of potential camera to projector correspondences. This is done by sweeping both a horizontal (row) and a vertical (column) scan line over the scene with a projector. This process is captured by a web cam on highly light sensitive settings. A depth consistency check is then performed to prune candidates that must have been lit indirectly because depths from potential matching rows match none of the depth from potential matching columns. The remaining candidates are scored on their combined scene continuity using dynamic programming with a cost function that rewards smoothness. To reduce the effect of subsurface scattering, adjacent candidates are grouped together and judged by their center. Additionally it is shown that multiple sweep lines can be run at once as the depth consistency check in combination with dynamic programming can solve most ambiguity. The proposed method was implemented in C++ and was tested on different scenes with indirect light against a structured light method specifically designed to handle indirect light. While the proposed method is a lot slower, it did outperformed the other method on resolution and accuracy and may therefore fill a niche of slow, but very cheap and highly accurate structured light methods.