Key learning features as means for terrain classification

I. Gheorghe, Weidong Li, T. Popham, A. Gaszczak, Keith Burnham

Research output: Chapter in Book/Report/Conference proceedingChapter

3 Citations (Scopus)

Abstract

Modern vehicles seek autonomous subsystems adaptability to everchanging terrain types in pursuit of enhanced drivability and maneuverability. The impact of key features on the classification accuracy of terrain types using a colour camera is investigated. A handpicked combination of texture and colour as well as a simple unsupervised feature representation is proposed. Although the results are restricted to only four classes {grass, tarmac, dirt, gravel} the learned features can be tailored to suit more classes as well as different scenarios altogether. The novel aspect stems from the feature representation itself as a global gist for three quantities of interest within each image: background, foreground and noise. In addition to that, the frequency affinity of the Gabor wavelet gist component to perspective images is mitigated by inverse homography mapping. The emphasis is thus on feature selection in an unsupervised manner and a framework for integrating learned features with standard off the shelf machine learning algorithms is provided. Starting with a colour hue and saturation histogram as fundamental building block, more complex features such as GLCM, k-means and GMM quantities are gradually added to observe their integrated effect on class prediction for three parallel regions of interest. The terrain classification problem is tackled with promising results using a forward facing camera.
Original languageEnglish
Title of host publicationAdvances in Intelligent Systems and Computing
EditorsJ. Swiątek, A. Grzech, P. Swiątek, J.M. Tomczak
PublisherSpringer Verlag
Pages273-282
VolumeAdvances in Intelligent Systems and Computing Volume 240
ISBN (Print)978-3-319-01856-0, 978-3-319-01857-7
DOIs
Publication statusPublished - 2014

Fingerprint

Color
Cameras
Maneuverability
Gravel
Learning algorithms
Learning systems
Feature extraction
Textures

Bibliographical note

The full text of this item is not available from the repository.

Keywords

  • Colour
  • Gist
  • GLCM
  • Homography
  • Machine learning
  • Terrain classification
  • Texture

Cite this

Gheorghe, I., Li, W., Popham, T., Gaszczak, A., & Burnham, K. (2014). Key learning features as means for terrain classification. In J. Swiątek, A. Grzech, P. Swiątek, & J. M. Tomczak (Eds.), Advances in Intelligent Systems and Computing (Vol. Advances in Intelligent Systems and Computing Volume 240, pp. 273-282). Springer Verlag. https://doi.org/10.1007/978-3-319-01857-7_26

Key learning features as means for terrain classification. / Gheorghe, I.; Li, Weidong; Popham, T.; Gaszczak, A.; Burnham, Keith.

Advances in Intelligent Systems and Computing. ed. / J. Swiątek; A. Grzech; P. Swiątek; J.M. Tomczak. Vol. Advances in Intelligent Systems and Computing Volume 240 Springer Verlag, 2014. p. 273-282.

Research output: Chapter in Book/Report/Conference proceedingChapter

Gheorghe, I, Li, W, Popham, T, Gaszczak, A & Burnham, K 2014, Key learning features as means for terrain classification. in J Swiątek, A Grzech, P Swiątek & JM Tomczak (eds), Advances in Intelligent Systems and Computing. vol. Advances in Intelligent Systems and Computing Volume 240, Springer Verlag, pp. 273-282. https://doi.org/10.1007/978-3-319-01857-7_26
Gheorghe I, Li W, Popham T, Gaszczak A, Burnham K. Key learning features as means for terrain classification. In Swiątek J, Grzech A, Swiątek P, Tomczak JM, editors, Advances in Intelligent Systems and Computing. Vol. Advances in Intelligent Systems and Computing Volume 240. Springer Verlag. 2014. p. 273-282 https://doi.org/10.1007/978-3-319-01857-7_26
Gheorghe, I. ; Li, Weidong ; Popham, T. ; Gaszczak, A. ; Burnham, Keith. / Key learning features as means for terrain classification. Advances in Intelligent Systems and Computing. editor / J. Swiątek ; A. Grzech ; P. Swiątek ; J.M. Tomczak. Vol. Advances in Intelligent Systems and Computing Volume 240 Springer Verlag, 2014. pp. 273-282
@inbook{6c9114192b8c43148a5c02dd4550b32d,
title = "Key learning features as means for terrain classification",
abstract = "Modern vehicles seek autonomous subsystems adaptability to everchanging terrain types in pursuit of enhanced drivability and maneuverability. The impact of key features on the classification accuracy of terrain types using a colour camera is investigated. A handpicked combination of texture and colour as well as a simple unsupervised feature representation is proposed. Although the results are restricted to only four classes {grass, tarmac, dirt, gravel} the learned features can be tailored to suit more classes as well as different scenarios altogether. The novel aspect stems from the feature representation itself as a global gist for three quantities of interest within each image: background, foreground and noise. In addition to that, the frequency affinity of the Gabor wavelet gist component to perspective images is mitigated by inverse homography mapping. The emphasis is thus on feature selection in an unsupervised manner and a framework for integrating learned features with standard off the shelf machine learning algorithms is provided. Starting with a colour hue and saturation histogram as fundamental building block, more complex features such as GLCM, k-means and GMM quantities are gradually added to observe their integrated effect on class prediction for three parallel regions of interest. The terrain classification problem is tackled with promising results using a forward facing camera.",
keywords = "Colour, Gist, GLCM, Homography, Machine learning, Terrain classification, Texture",
author = "I. Gheorghe and Weidong Li and T. Popham and A. Gaszczak and Keith Burnham",
note = "The full text of this item is not available from the repository.",
year = "2014",
doi = "10.1007/978-3-319-01857-7_26",
language = "English",
isbn = "978-3-319-01856-0",
volume = "Advances in Intelligent Systems and Computing Volume 240",
pages = "273--282",
editor = "J. Swiątek and A. Grzech and P. Swiątek and J.M. Tomczak",
booktitle = "Advances in Intelligent Systems and Computing",
publisher = "Springer Verlag",
address = "Austria",

}

TY - CHAP

T1 - Key learning features as means for terrain classification

AU - Gheorghe, I.

AU - Li, Weidong

AU - Popham, T.

AU - Gaszczak, A.

AU - Burnham, Keith

N1 - The full text of this item is not available from the repository.

PY - 2014

Y1 - 2014

N2 - Modern vehicles seek autonomous subsystems adaptability to everchanging terrain types in pursuit of enhanced drivability and maneuverability. The impact of key features on the classification accuracy of terrain types using a colour camera is investigated. A handpicked combination of texture and colour as well as a simple unsupervised feature representation is proposed. Although the results are restricted to only four classes {grass, tarmac, dirt, gravel} the learned features can be tailored to suit more classes as well as different scenarios altogether. The novel aspect stems from the feature representation itself as a global gist for three quantities of interest within each image: background, foreground and noise. In addition to that, the frequency affinity of the Gabor wavelet gist component to perspective images is mitigated by inverse homography mapping. The emphasis is thus on feature selection in an unsupervised manner and a framework for integrating learned features with standard off the shelf machine learning algorithms is provided. Starting with a colour hue and saturation histogram as fundamental building block, more complex features such as GLCM, k-means and GMM quantities are gradually added to observe their integrated effect on class prediction for three parallel regions of interest. The terrain classification problem is tackled with promising results using a forward facing camera.

AB - Modern vehicles seek autonomous subsystems adaptability to everchanging terrain types in pursuit of enhanced drivability and maneuverability. The impact of key features on the classification accuracy of terrain types using a colour camera is investigated. A handpicked combination of texture and colour as well as a simple unsupervised feature representation is proposed. Although the results are restricted to only four classes {grass, tarmac, dirt, gravel} the learned features can be tailored to suit more classes as well as different scenarios altogether. The novel aspect stems from the feature representation itself as a global gist for three quantities of interest within each image: background, foreground and noise. In addition to that, the frequency affinity of the Gabor wavelet gist component to perspective images is mitigated by inverse homography mapping. The emphasis is thus on feature selection in an unsupervised manner and a framework for integrating learned features with standard off the shelf machine learning algorithms is provided. Starting with a colour hue and saturation histogram as fundamental building block, more complex features such as GLCM, k-means and GMM quantities are gradually added to observe their integrated effect on class prediction for three parallel regions of interest. The terrain classification problem is tackled with promising results using a forward facing camera.

KW - Colour

KW - Gist

KW - GLCM

KW - Homography

KW - Machine learning

KW - Terrain classification

KW - Texture

U2 - 10.1007/978-3-319-01857-7_26

DO - 10.1007/978-3-319-01857-7_26

M3 - Chapter

SN - 978-3-319-01856-0

SN - 978-3-319-01857-7

VL - Advances in Intelligent Systems and Computing Volume 240

SP - 273

EP - 282

BT - Advances in Intelligent Systems and Computing

A2 - Swiątek, J.

A2 - Grzech, A.

A2 - Swiątek, P.

A2 - Tomczak, J.M.

PB - Springer Verlag

ER -