Abstract:
This paper presents a novel strategy for the construction of
dense three-dimensional environment models by combining
images from a conventional camera and a range imager.
Robust data association is first accomplished by exploiting
the Scale Invariant Feature Transformation (SIFT)
technique on the textured images. The two-dimensional
feature locations are then identified in the range images
and the full 3D information is in turn employed to calculate
the relative registration between consecutive camera poses.
Finally, the information from the range images is combined
with the newly obtained transformation to form a single
model of the environment. Validating results from a robot
operating in a rescue scenario are presented. This is a
typical scenario where the technique can be most useful as
robot odometry, if present, is often highly unreliable and
other means of locating the robot are necessary.
