Abstract

We present a comparison of several local image descriptors in the context of visual
Simultaneous Localization and Mapping (SLAM). In visual SLAM a set of points in the
environment are extracted from images and used as landmarks. The points are represented
by local descriptors used to resolve the association between landmarks. In this paper, we
study the class separability of several descriptors under changes in viewpoint and scale.
Several experiments were carried out using sequences of images in 2D and 3D scenes.