Fusion of 2D and 3D Modeling and Recognition Algorithms based on Multi-Modal Sensor Input

Structure from Motion (SfM) is a technique that estimates 3D structure given a 2D image sequence and is well suited to reconstruct the overall shape. It depends on finding feature correspondences in the images, therefor enough texture is necessary.
Photometric Stereo (PS) is a technique for estimating the photometric normals of objects by observing that object under varying lighting conditions. This technique is well suited to recover the details in the structure, but can introduce bias in the lower frequencies.
This thesis researches the combination of both methods, in a multiview setting with multiple lights. The goal of this thesis is to develop a method for ESAT’s minidome that uses both techniques to obtain more accurate 3D models, not only for diffuse materials, but also for highly specular materials.