| A key component of a mobile
robot system is the ability to localize itself accurately and, simultaneously,
to build a map of the environment. Most of the existing algorithms are based
on laser range finders, sonar sensors or artificial landmarks. In this paper,
we describe a vision-based mobile robot localization and mapping algorithm,
which uses scale-invariant image features as natural landmarks in unmodified
environments. The invariance of these features to image translation, scaling
and rotation makes them suitable landmarks for mobile robot localization
and map building. With our Triclops stereo vision system, these landmarks
are localized and robot ego-motion is estimated by least-squares minimization
of the matched landmarks. Feature viewpoint variation and occlusion are
taken into account by maintaining a view direction for each landmark. Experiments
show that these visual landmarks are robustly matched, robot pose is estimated
and a consistent three-dimensional map is built. As image features are not
noise-free, we carry out error analysis for the landmark positions and the
robot pose. We use Kalman filters to track these landmarks in a dynamic
environment, resulting in a database map with landmark positional uncertainty. |
