Publication Details

Reducing the Run-time Complexity of Support Vector Machine Used for Rail Candidates Detection

MUSIL, M. Reducing the Run-time Complexity of Support Vector Machine Used for Rail Candidates Detection. International Masaryk conference for Ph.D. students and young researchers. vol. VI. Hradec Králové: Akademické sdružení MAGNANIMITAS Assn., 2015. p. 2138-2146. ISBN: 978-80-87952-12-2.
Czech title
Snížení časové složitosti Support Vector Machine pro detekci kandidátů na kolejnice
Type
conference paper
Language
English
Authors
Musil Marek, Ing.
URL
Keywords

computer-vision, Histogram of Oriented Gradients (HOG), optimization, performance, rail candidates detection, run-time complexity, Support Vector Machine (SVM)

Abstract

Support Vector Machine (SVM) is a technique for classification and regression. It uses a decision surface called hyperplane that depends on the regularization parameter and training points lying in the margin of the hyperplane. The run-time complexity of SVM may be reduced through the hyperplane affected by the regularization parameter. We deal with rails recognition in images taken from the camera mounted on the board of the locomotive. For the purpose of rail candidates detection, we deployed an algorithm using SVM. We performed several experiments under different settings. In this paper, we introduce an algorithm using SVM and the impact of its regulation parameter as well as others possible on SVM-performance. The main goal is to decrease time-complexity while maintaining classification success rate.

Published
2015
Pages
2138–2146
Proceedings
International Masaryk conference for Ph.D. students and young researchers
Series
vol. VI
ISBN
978-80-87952-12-2
Publisher
Akademické sdružení MAGNANIMITAS Assn.
Place
Hradec Králové
BibTeX
@inproceedings{BUT123624,
  author="Marek {Musil}",
  title="Reducing the Run-time Complexity of Support Vector Machine Used for Rail Candidates Detection",
  booktitle="International Masaryk conference for Ph.D. students and young researchers",
  year="2015",
  series="vol. VI",
  pages="2138--2146",
  publisher="Akademické sdružení MAGNANIMITAS Assn.",
  address="Hradec Králové",
  isbn="978-80-87952-12-2",
  url="http://www.vedeckekonference.cz/index.php?option=com_content&view=article&id=79&Itemid=66&lang=en"
}
Back to top