master thesis

SignatureMA 015
AuthorPeter Stihl
TitleImplementation and Evaluation of Surface Reconstruction from Unstructured Point Clouds
TutorDipl.-Inf. Thilo Borgmann
ProfessorProf. Dr.-Ing. Thomas Sikora
AbstractThe video or image content for modern 3D-display-devices requires at least two perspective views according to the two eyey of the human vision system. A common approach in production of such footage is the synthetic rendering of a computer model representing the desired scene from the appropriate view points. Image-based stereo reconstruction methods generate an unstructured point cloud describing the surface of the scene. Through surface reconstruction algorithms these point clouds can be transferred into a polygonal mesh that can be synthesized via rendering techniques from the field of Computer Graphics.
This work aims for an evaluation whether a combination of multiple point clouds, each generated by two-frame stereo reconstruction algorithms, allows for a more accurate surface to be reconstructed rather than considering only one image pair.
Different surface reconstruction algorithms for transferring unstructured point clouds into triangular meshes are implemented and universally evaluated in this work. The specific evaluation against the above stated setup shows that the unmodified algorithms are rather unsuited for this context.
Key wordssurface reconstruction, triangulation, point clouds, implementation, evaluation, Computer Graphics, 3D