Investigation of Alternative Contactless Optical Surface Reconstruction Methods

Abstract: Object and surface reconstruction in 3D is commonly used in production tolerance validation. Depending on the required level of accuracy, a multitude of different methods are available, each with their own advantages and disadvantages. This thesis aims to investigate alternative surface reconstruction methods which utilise that all possible light paths are known along a given one-dimensional curve on the surface of a material with known refractive index. This makes it possible to calculate the expected light intensity for the curve. By comparing the expected intensity to a reference intensity from when there is no surface to reflect off, it is possible to deduce information about the surface. Several methods utilising this information to reconstruct the surface of an object are investigated and evaluated. The types of investigated methods range from one-dimensional iterative methods to Convolutional Neural Network based Encoder-Decoder architectures. The evaluation shows that it is possible to deduce information about the general shape of the surface, but that the non-linear nature of the problem makes it difficult to identify any fine details. The methods showing the most promising results use a combination of principal components and simple geometric relationships in the data to reconstruct the surface.