Multi-scale space smoothing and segmentation of range data for robot navigation
One of the most essential problems for mobile robot navigation is to enable an autonomous robot to navigate in an unknown environment and to incrementally build a map of this environment while simultaneously using this map to compute its current location. This problem is usually referred to as Simul...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Theses and Dissertations |
Language: | English |
Published: |
2010
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/42278 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
id |
sg-ntu-dr.10356-42278 |
---|---|
record_format |
dspace |
spelling |
sg-ntu-dr.10356-422782023-07-04T16:13:23Z Multi-scale space smoothing and segmentation of range data for robot navigation Tang, Fan Martin David Adams Wijerupage Sardha Wijesoma School of Electrical and Electronic Engineering Robotics Research Centre DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics One of the most essential problems for mobile robot navigation is to enable an autonomous robot to navigate in an unknown environment and to incrementally build a map of this environment while simultaneously using this map to compute its current location. This problem is usually referred to as Simultaneous Localization and Mapping (SLAM). The feature-based SLAM approach has become one of the most promising solutions due to the tractability of its map. However it requires a robust method to extract enough robot pose invariant detectable landmarks from the surrounding environment. This thesis focuses on feature extraction methods from range data based on several filtering algorithms. Firstly, the bilateral filtering algorithm based on Kalman filter is proposed to extract robot pose invariant features from range data. With the ideas of multi-scale filtering from image processing, an adaptive smoothing algorithm, with a model based mask, within a scale space framework is proposed for feature extraction from range data. This algorithm smoothes range data and segments it at the same time by translating a model based mask over the data. The weights of the smoothing mask are adaptively calculated according to the Mahalanobis distance between range data and model based predictions. The model based mask smoothing technique with adaptive weights is applied in multi-scale space. The convergence of the algorithm is also proved in terms of its compliance with the anisotropic diffusion concept from the vision literature. Thus, more robust and pose-invariant features can be extracted. All the filtering algorithms have been implemented and verified with both simulated range data and real data collected from a laser range sensor mounted on a robot vehicle. In order to assess the improvement in SLAM by using those features extracted by the proposed filtering algorithms, a full SLAM algorithm is implemented. By comparing the estimated robot location and map, it will be shown that the features extracted by the proposed filtering algorithms are robot pose invariant and have higher location accuracy than the features extracted by traditional methods. DOCTOR OF PHILOSOPHY (EEE) 2010-10-14T08:07:32Z 2010-10-14T08:07:32Z 2010 2010 Thesis Tang, F. (2010). Multi-scale space smoothing and segmentation of range data for robot navigation. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/42278 10.32657/10356/42278 en 208 p. application/pdf |
institution |
Nanyang Technological University |
building |
NTU Library |
continent |
Asia |
country |
Singapore Singapore |
content_provider |
NTU Library |
collection |
DR-NTU |
language |
English |
topic |
DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics |
spellingShingle |
DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics Tang, Fan Multi-scale space smoothing and segmentation of range data for robot navigation |
description |
One of the most essential problems for mobile robot navigation is to enable an autonomous robot to navigate in an unknown environment and to incrementally build a map of this environment while simultaneously using this map to compute its current location. This problem is usually referred to as Simultaneous Localization and Mapping (SLAM). The feature-based SLAM approach has become one of the most promising solutions due to the tractability of its map. However it requires a robust method to extract enough robot pose invariant detectable landmarks from the surrounding environment. This thesis focuses on feature extraction methods from range data based on several filtering algorithms. Firstly, the bilateral filtering algorithm based on Kalman filter is proposed to extract robot pose invariant features from range data. With the ideas of multi-scale filtering from image processing, an adaptive smoothing algorithm, with a model based mask, within a scale space framework is proposed for feature extraction from range data. This algorithm smoothes range data and segments it at the same time by translating a model based mask over the data. The weights of the smoothing mask are adaptively calculated according to the Mahalanobis distance between range data and model based predictions. The model based mask smoothing technique with adaptive weights is applied in multi-scale space. The convergence of the algorithm is also proved in terms of its compliance with the anisotropic diffusion concept from the vision literature. Thus, more robust and pose-invariant features can be extracted.
All the filtering algorithms have been implemented and verified with both simulated range data and real data collected from a laser range sensor mounted on a robot vehicle. In order to assess the improvement in SLAM by using those features extracted by the proposed filtering algorithms, a full SLAM algorithm is implemented. By comparing the estimated robot location and map, it will be shown that the features extracted by the proposed filtering algorithms are robot pose invariant and have higher location accuracy than the features extracted by traditional methods. |
author2 |
Martin David Adams |
author_facet |
Martin David Adams Tang, Fan |
format |
Theses and Dissertations |
author |
Tang, Fan |
author_sort |
Tang, Fan |
title |
Multi-scale space smoothing and segmentation of range data for robot navigation |
title_short |
Multi-scale space smoothing and segmentation of range data for robot navigation |
title_full |
Multi-scale space smoothing and segmentation of range data for robot navigation |
title_fullStr |
Multi-scale space smoothing and segmentation of range data for robot navigation |
title_full_unstemmed |
Multi-scale space smoothing and segmentation of range data for robot navigation |
title_sort |
multi-scale space smoothing and segmentation of range data for robot navigation |
publishDate |
2010 |
url |
https://hdl.handle.net/10356/42278 |
_version_ |
1772828557543735296 |