Kristoffer Sjö,
Dorian Galvez Lopez, Chandana Paul,
Patric Jensfelt and
Danica Kragic
Abstract:
In this paper we present a method for search and localization of
objects with a mobile robot using a monocular camera with zoom
capabilities. We show how to overcome the limitations of low
resolution images in object recognition by utilizing a combination of
an attention mechanism and zooming as the first steps in the
recognition process. The attention mechanism is based on receptive
field cooccurrence histograms and the object recognition on SIFT
feature matching. We present two methods for estimating the distance
to the objects which serves both as the input to the control of the
zoom and the final object localization. Through extensive experiments
in a realistic environment, we highlight the strengths and weaknesses
of both methods. To evaluate the usefulness of the method we also
present results from experiments with an integrated system where a
global sensing plan is generated based on view planning to let the
camera cover the space on a per room basis.
BibTeX Entry:
Download: pdf (494k)