Target localization and identification using CTFM sonar imaging: the AURBIT method

Abstract

We introduce a new method for acquiring and processing ultrasound signals for the location and identification of the typical primitives for map building - planes, corners and edges. Models of the echo signals received allows the feature parametrisation of their range and amplitude values. Part of the pre-processing and the classification over their standard features was done by using neural networks. Receiver saturation due to the large dynamic range is included in the models. The combination of model based methods and learning provides an algorithm which requires only sparse measurements to locate and identify targets. The transducer is a continuous wave single frequency modulated (CFTM) transmitter/receiver, not original in terms of its hardware but with little reported application in this field. The method is applied to mobile robot exploration and demonstrated with the production of a sonar map.