| |
ABSTRACT
A mobile robot needs to perceive the motions of
external objects to perform tasks successfully in a dynamic
environment. We propose a set of algorithms for multiple motion
tracking from a mobile robot equipped with a monocular camera
and a laser rangefinder. The key challenges are 1. to compensate
the ego-motion of the robot for external motion detection, and 2.
to cope with transient and structural noise for robust motion
tracking. In our algorithms, the robot ego-motion is directly
estimated using corresponding feature sets in two consecutive
images, and the position and velocity of a moving object is
estimated in image space using multiple particle filters. The
estimates are fused with the depth information from the laser
rangefinder to estimate the partial 3D position. The proposed
algorithms have been tested with various configurations in
outdoor environments. The algorithms were deployed on three
different platforms; it was shown that various type of ego-motion
were successfully eliminated and the particle filter was able to
track motions robustly. The multiple target tracking algorithm
was tested for different types of motions, and it was shown that
our multiple lter approach is effective and robust. The tracking
algorithm was integrated with a robot control loop, and its realtime
capability was demonstrated.
|
