A comparison of two methods for natural landmark classification with Biosonar

Abstract

Most current sonar systems for mobile robots only yield time of flight distance information produced by measuring the time of a single sonar pulse. However, sonar systems of animals like sonar bats are much more sophisticated, allowing to recognize not only the shape, but also the type of landmark trees that they use during their nocturnal flights. In this paper we compare two methods for natural landmark classification by a biomimetic sonar consisting of one sender (mouth) and two receivers (ears). Our research aims to improve the navigation capabilities of mobile robots in natural and man-made environments, where economical and robust feature extrcation of potential landmarks such as trees is a key problem. We try to classify different species of trees from random orientations. Two methods utilized are intricate structure feature based and general structure feature based natural landmark classification. After a comparation of those two method's advantages and disadvantages, we forwarded a two-step Biosonar classification strategy for outdoor navigation of mobile robots. Experimental results indicate that a mobile robot with such a Biosonar system can achieve the ability to classify natural landmarks like trees only based on sonar.