A Statistical Learning Method for Mobile Robot Environment Modeling

Abstract

This paper presents a statistical learning method for computing range data
as an initial solution to the environment modeling problem in the context
of mobile robotics. Unlike other methods that are based on a set of geometric
primitives, our method computes dense range maps of locations in the
environment using only intensity images and very limited amount of range
data as an input. This is achieved by exploiting the following assumptions:
1) the observed range and intensity images are correlated and, 2) variations
of pixels in the range and intensity images are related to the values
elsewhere in the image(s). These variations can be efficiently captured by
the neighborhood system of a Markov Random Field (MRF). Experimental results
show the feasibility of our method.