Spatial inference – Learning vs. constraint solving

Abstract

We present a comparison of two new approaches for solving constraints occurring in spatial inference. In contrast to qualitative spatial reasoning we use a metric description, where relations between pairs of objects are represented by parameterized homogenous transformation matrices with numerical (nonlinear) constraints. We employ interval arithmetics based constraint solving and methods of machine learning in combination with a new algorithm for generating depictions for spatial inference.