Correspondence detection in natural terrain : computer vision techniques for noisy 3D range data
Abstract/Contents
- Abstract
- Exploration of small bodies such as asteroids and comets can provide valuable scientific understanding of the history and evolution of our solar system, and a key component in these missions is generating a shape model of the target body. LiDAR systems are an emerging option for small body mapping, but face challenges with the point cloud correspondence detection step of the pipeline. Correlation-based correspondence detection techniques require a good initial alignment in order to reliably converge to an accurate result, and the high uncertainty of small bodies' motion models, coupled with the lack of external navigational infrastructure such as GNSS or fixed beacons makes this a challenging problem. With varying range sensor resolutions, high noise levels, and low-gradient terrain, the signal to noise level of the point cloud can cause feature-based techniques to have a low feature recall and a high ratio of outliers. When the overall number of features is small, which can be common in natural terrain applications and as is expected near safe rover landing sites, state of the art feature-based correspondence methods can unknowingly generate misaligned results. In this thesis, a novel point cloud correspondence detection algorithm is presented that combines constellations of a small number of features within point cloud data with Iterative Closest Point (ICP) alignment solutions. First, a new handcrafted 3D point cloud descriptor, Smooth Signature of Histograms of OrienTations (SmoothSHOT), is tailored to natural terrain. Next, the Pyramid star tracker constellation detection algorithm is fused with the MinBucket graph matching algorithm to provide an efficient database search procedure, Pyramid Edge-Reduction Schema for Efficient Isomorphism Detection (PERSEID). Constellations of features are combined into metafeatures in order to identify the most distinctive constellation matches within the dataset and prevent susceptibility to repetitive or ambiguous geometry. Finally, ICP is initialized with the metafeature alignment, and then the result is validated by it as well. When applied to a model of the asteroid Itokawa with realistic noise and resolution changes, ICP alone achieves a precision of only 60-70% across a range of noise conditions and resolution changes. RANdom SAmple and Consensus (RANSAC) initialized ICP has 99% precision at low noise settings, but only 44% at the realistic noise conditions. The algorithm presented in this thesis, PERSEID-Augmented ICP (PAI), is able to achieve a correspondence detection precision of 100% under the same conditions. On an underwater sonar dataset where ICP and RANSAC-initialized ICP have 53% and 32% precision, respectively, the PAI algorithm is able to achieve 90% precision. Higher-precision correspondence detection results with fewer required feature matches can improve point cloud correspondence detection for extraterrestrial mapping in feature-poor areas.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2024; ©2024 |
Publication date | 2024; 2024 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Clark, Ashley A. (Ashley Anne) |
---|---|
Degree supervisor | Rock, Stephen M |
Thesis advisor | Rock, Stephen M |
Thesis advisor | D'Amico, Simone |
Thesis advisor | Schwager, Mac |
Degree committee member | D'Amico, Simone |
Degree committee member | Schwager, Mac |
Associated with | Stanford University, School of Engineering |
Associated with | Stanford University, Department of Aeronautics and Astronautics |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Ashley A. Clark. |
---|---|
Note | Submitted to the Department of Aeronautics and Astronautics. |
Thesis | Thesis Ph.D. Stanford University 2024. |
Location | https://purl.stanford.edu/gc656cn5590 |
Access conditions
- Copyright
- © 2024 by Ashley Anne Clark
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...