Experiments in robust bistatic sonar object classification for local environment mapping

Abstract

Presents the classification results of a bistatic sonar sensor with decision tree classifier for use in mobile robot navigation. Unlike previous work the paper investigates the discrimination and robustness of the sensor's classifications when presented with common office objects with complex geometries. The feature extraction process uses a novel perturbed L/sub 2/ method which allows a physical interpretation of the features. The robustness of the classifications indicate that, given a suitably enlarged set of training objects, the approach would be suitable for use within office environments.