Unsupervised Detection of Artificial Objects in Outdoor Environments
Résumé
This paper presents a novel unsupervised sensor fusion method to detect artificial objects in outdoor environments. We define artificial objects present in outdoor environments using structure and appearance information: an artificial object is composed of several smooth surfaces with sufficiently extended area and distinctive colors with respect to the environment. Structure information is obtained extracting smooth linked surfaces from 3D range data. Appearance information is computed on image data through a color processing. The problem of fusing these two different kinds of information is addressed through the use of a Bayesian sensor fusion approach. A probability map is built and then clustered with a 2.5D unsupervised self organizing network. This method defines objects according to the distribution of the probability values in the fusion map. The resulting clusters are labeled with their mean probability value representing the confidence the detected regions have of being artificial objects. In order to show the validity of the proposed algorithm some experiments have been performed in real outdoor environments showing promising results.
Domaines
Robotique [cs.RO]Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...