3D Vision
Global Alignment of Meshes for the Microsoft HoloLens

During my semester abroad at ETH, I had a chance to work on a research project in 3D Vision under the Computer Vision and Geometry Group with two other students in developing a mesh registration pipeline with Guaranteed Outlier Removal (GORE) and other refinement techniquess.

01 Abstract

With devices such as the Microsoft HoloLens now able to scan and record 3D scenes as meshes, a novel way of registering scenes taken from different coordinate systems and at different times is required.

We base our work off of the Guaranteed Outlier Removal Method presented by Bustos and Chin and expand it beyond point clouds to analyze 3D meshes. However, we could not conclude that GORE can perform consistently and effectively on real-world datasets generated by the HoloLens.

In particular, we identified that GORE has the most difficulty when the points from two datasets do not have significant overlap, notably any values below the 80% threshold. This places severe limitations on the applications of GORE, especially on real-life alignment problems.

In a team of 3, this semester-long research project supervised under Dr. Martin Oswald resulted in a final paper, presentation, and poster session with the Computer Vision and Geometry Group at ETH.