A Survey on Point Cloud Registration

PhD Qualifying Examination


Title: "A Survey on Point Cloud Registration"

by

Mr. Xuyang BAI


Abstract:

Point cloud registration is a well-studied area in the computer vision
community, especially with the prevalence of affordable 3D scanners, RGB-D
cameras, and LiDARs. Given two partially overlapped point clouds, the aim
is to find the optimal transformation between them. It is a fundamental
stone of computer vision applications such as simultaneous localization
and mapping (SLAM) and 3D LiDAR-based mapping. However, it is a
challenging task due to the irregularity and sparsity of the 3D point
cloud, which prevents the adoption of existing neural networks. Thus, new
network architectures and processing strategies are necessary to {aid} the
registration of two different point clouds. Also, additional efforts on
algorithm robustness have to be paid in order to align the point clouds
under noisy and low-overlapping situations.

In this survey, we broadly classify the existing point cloud registration
methods into two categories: coarse registration and fine registration,
and review the related techniques in each category. For coarse
registration, we introduce the methods that solve each step of the coarse
registration pipeline by discussing a few representative works. For fine
registration, we mainly discuss the ICP family and NDT family. We conclude
by discussing the current challenges of point cloud registration and
providing two interesting research directions for future work.


Date:                   Monday, 24 February 2020

Time:                   4:30pm - 6:30pm

Zoom Meeting:           https://hkust.zoom.us/j/3966929732

Committee Members:      Prof. Chiew-Lan Tai (Supervisor)
                        Prof. Chi-Keung Tang (Chairperson)
                        Prof. Huamin Qu
                        Dr. Pedro Sander


**** ALL are Welcome ****