Shape from colored structured light and polarization

Abstract

In this thesis data from two separate methods, shape from diffuse polarization and rapid shape acquisition using color structured light are combined into a single point cloud. Several improvements to the structured light method are proposed which decrease the negative impact of textured surfaces and enable sub-pixel accuracy. The improvements of the structured light method are tested by comparison with a ground truth model. The results fall within a 0.8 mm standard deviation of the ground truth, with outliers to 0,6 mm accuracy for some models when all improvements are used. The accuracy of the polarized light method is also tested in several situations. The accuracy of this method increases in areas that are almost parallel to the light direction, leading up to 30% less noise when compared to areas where the light is perpendicular to the light direction. The combination of the methods results in a ten times higher resolution pointcloud than would be possible with only the rapid shape acquisition. Unfortunately, there is a signi?cant increase in noise as compared to the unmodified method, because if one of the methods provides incorrect data, the combination propagates these mistakes. Better results should be obtainable with a different gradient to depth algorithm and multiple viewpoints for the polarized light. The implemented system is modular, enabling easy modification and switching of methods.