A Robust Feature Matching Method for Robot Localization in a Dynamic Indoor Environment - Technologies and Applications of Artificial Intelligence - page 352

A Robust Feature Matching Method for Robot
Localization in a Dynamic Indoor Environment

Tsung-Yen Tsou and Shih-Hung Wu *

Department of Computer Science and Information Engineering,
Chaoyang University of Technology, Taichung City 41349, Taiwan (R.O.C)
[e-mail], [e-mail]

Abstract. In this paper, we report how the feature matching method can be
applied to deal with the indoor mobile robot localization problem. We assume
that a robot equipped with a laser rangefinder can scan the environment in real
time and get the geometry features, and then the robot can match these features
with those collected in advance to find the possible locations. This approach
would face two difficulties. Since there are locations with similar features, the
robot have to move around and do the scan and match several times to make sure
the right location. There is another difficult problem, the features might not be fix
in real-world dynamic environment, e.g. people might be walking through,
furniture might be shifted; therefore, a robust feature matching method is needed
for dynamic environment. This paper describes an efficient method using
omni-directional feature grouping to improve the feature matching method for
robot localization. With the laser rangefinder, a robot finds the 360 degree
coverage information. Omni-directional feature grouping has the advantage of
dividing all the features of a hypothetical position through different directions to
generate multiple sets of environmental features. The method can reduce the
affection of moving objects in a dynamic environment. Experimental results
show that our method improve the accuracy rate and has low average errors.

Keywords: Indoor localization, laser rangefinder, feature matching, robot
navigation, dynamic indoor environment

1 Introduction

Indoor localization is an important issue for mobile robot research. There are many
previous works focusing on the simultaneous localization and map building (SLAM)
problem [1,2,3,4], [18], [22], which combine map building and localization as
one problem. Unlike outdoor environment, where GPS can provide mobile robot
reliable location information, GPS is not available in indoor environment. In order to
deal with this difficulty, various sensors have been used in indoor localization, such as
laser rangefinder, sonar, odometer, and camera.
There are two types of indoor localization problems. First one is the robot knows
the initial location. This type is easier since robot must be in the neighborhood of the
* Corresponding author.