Back to Top

Indoor Place Recognition Using Support Vector Machines

A. Pronobis

Master's thesis, KTH Royal Institute of Technology, 2005.


The ability to recognize places on the basis of visual perceptual information is a fundamental property of human beings, and thus determines the way we communicate and alter our surroundings. Consequently, it becomes indispensable to provide similar capabilities for machines aiming to interact with humans and man-made environment. In this thesis we address the problem of visual indoor place recognition. We propose a solution based on Support Vector Machines employing both global and local image descriptors. Since robustness and efficiency are crucial for every recognition system aiming to work in real-world settings, we put special emphasis on these properties. We build a database comprising several sets of pictures acquired in five rooms of different functionality, under various conditions. We then use it in order to evaluate the performance of our system, and achieve very good results in presence of variations that occur in real environments. Additionally, for sake of efficiency, we implement an algorithm allowing for an exact simplification of support vector solutions. We further extend the original algorithm so that it could provide higher efficiency gain by means of approximation. The results reported in the thesis show great potential of our method in a wide range of computer vision applications and prove that support vector solutions can be successfully applied to the place recognition problems.


  author =       {Pronobis, Andrzej},
  title =        {Indoor Place Recognition Using {S}upport {V}ector {M}achines},
  year =         2005,
  month =        dec,
  school =       {KTH Royal Institute of Technology},
  address =      {Stockholm, Sweden},
  url =          {}
© 2018. Copyright Andrzej Pronobis
  • stackoverflow
  • scholar