Title: Learning for Registration in 2D and 3D

Abstract: We explore the application of deep learning to 2D (image) and 3D (point cloud) registration, especially in scenarios where traditional methods can fail.

In the 2D case, we apply a recently-proposed learning method to the problem of aligning outdoor imagery taken across different seasons or times of day. Further, we extend the method to perform GPS-denied UAV geolocalization by aligning UAV images and satellite imagery.

In the 3D case, we develop three novel point cloud registration algorithms based on state-of-the-art networks for point cloud processing. We show the benefits of the learned representation in terms of robustness to initialization, noisy data, object generalizability, and speed.

Committee:

Simon Lucey (Advisor)

Michael Kaess

Leo Keselman