Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models
Thesis (MEng)--Stellenbosch University, 2024.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613918433640448 |
|---|---|
| access_status_str | Open Access |
| author | Dorfling, Anchal |
| author2 | Van Daalen, C. E. |
| author_browse | Dorfling, Anchal Van Daalen, C. E. |
| author_facet | Van Daalen, C. E. Dorfling, Anchal |
| author_sort | Dorfling, Anchal |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2024. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/131644 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:43:46.817Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/131644 Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models Dorfling, Anchal Van Daalen, C. E. Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Robotics Computer vision Robots -- Control systems Probabilistic automata Visual odometry Motion segmentation UCTD Thesis (MEng)--Stellenbosch University, 2024. ENGLISH ABSTRACT: In autonomous navigation, a robot uses features (or points of interest) on static objects in the environment to estimate its own motion, also known as visual odometry. However, in a dynamic environment, features on static objects (or, static features) must first be identified before egomotion estimation can take place. Classifying features on objects as static or dynamic is known as motion segmentation; for this, knowledge of the robot’s motion is required. Furthermore, measurements of features on static objects may not always be correct due to errors in the measurement extraction process; such measurements are termed as outliers. Consequently, outlier detection must be performed as to not feed the visual odometry process with incorrect measurements. However, the knowledge of the robot’s motion is required to identify outliers and inliers (or correct measurements). Thus, there is a dependency between visual odometry, motion segmentation and outlier detection; consequently, they must be performed simultaneously. Existing research in this field is scarce: current methods either make assumptions about the robot or the environment, or use high-end equipment for computation. This limits the applicability of the method, specially for applications where generic, low-cost solutions are desired. This thesis proposes a probabilistic graphical model (PGM) based approach to develop a flexible, low-cost solution for simultaneous robust visual odometry (that is, visual odometry and outlier detection) and motion segmentation. PGMs provide a framework for encoding uncertainty of quantities as well as their statistical dependencies. During the modelling phase of using PGMs, hybrid factors and non-linearities are encountered which complicates the inference phase. To address the issue of hybrid factors, the expectation propagation belief update algorithm is used to perform inference while linearisation techniques are applied to handle the non-linear relationships. Experimental results indicate that due to interdependencies present in the problem and naive initialisation of inference, the PGM performs poorly. To improve performance, a separate, supplementary PGM is designed and implemented to initialise inference on the main PGM (that is designed to solve simultaneous robust visual odometry and motion segmentation). The supplementary PGM (that is, the initialising PGM) aims to provide good starting points for inference on the main PGM in a quick and cheap way. Additional experimental results indicate that good initialisation significantly improves the performance of the main PGM. This initialising PGM is considered to be novel as it does not require the knowledge of previously found solutions nor is it computationally expensive to implement (as opposed to existing methods). The developed solution (that consists of both the initialising and the main PGM) was applied to the KITTI datasets where results indicate that a PGM approach works well and is a promising method to solve simultaneous robust visual odometry and motion segmentation. AFRIKAANSE OPSOMMING: Geen opsomming beskikbaar. Masters 2025-02-03T09:49:05Z 2025-02-03T09:49:05Z 2024-12 Thesis https://scholar.sun.ac.za/handle/10019.1/131644 en Stellenbosch University xiii, 93 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Robotics Computer vision Robots -- Control systems Probabilistic automata Visual odometry Motion segmentation UCTD Dorfling, Anchal Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| title | Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| title_full | Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| title_fullStr | Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| title_full_unstemmed | Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| title_short | Simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| title_sort | simultaneous robust visual odometry and motion segmentation for mobile robots using probabilistic graphical models |
| topic | Robotics Computer vision Robots -- Control systems Probabilistic automata Visual odometry Motion segmentation UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/131644 |
| work_keys_str_mv | AT dorflinganchal simultaneousrobustvisualodometryandmotionsegmentationformobilerobotsusingprobabilisticgraphicalmodels |