Henrik Andreasson and Tom Duckett
Topological Localization for Mobile Robots using Omni-directional Vision and Local Features
Proc. of the 5th IFAC Symposium on Intelligent Autonomous Vehicles (IAV), 2004, pp. 3-11
Abstract: This paper proposes a new method for recognizing typical objects found in indoor office environments (tables, chairs, etc.,) by a mobile robot equipped with an omni-directional vision sensor, without requiring any pre-installed geometric models of objects. The approach utilizes the motion of the robot to acquire an internal representation of a given object using ``structure from motion'' or optic flow. First, a set of low-level point features are selected from the segmented area of the image containing the object. The low-level features are tracked by a set of independent Kalman filters as the robot moves through the environment, in order to extract the 3D positions of these points. A set of high-level features is then extracted for input to a pattern recognition system, based on the spatial distribution of the low-level point features. The same feature extraction method is then applied for recognition of the learned objects. Results are presented for some first experiments on a real robot in a laboratory environment.
Paper: [PDF (3.4MB)]
  AUTHOR = {Andreasson, Henrik and Duckett, Tom},
  TITLE = {Topological Localization for Mobile Robots Using Omni-directional Vision and Local Features},
  BOOKTITLE = {Proc. of the 5th IFAC Symposium on Intelligent Autonomous Vehicles (IAV)},
  YEAR = {2004},
  DATE = {July, 5 -- 7},
  ADRESS = {Lisbon, Portugal}