Three-dimensional Real-time Localization of Subsurface Objects – from Theory to Experimental Validation

In the last years, significant efforts have been made to develop unsupervised systems able to detect landmines or unexploded ordnances for both military and civilian purposes. Several solutions have been proposed based on different methodologies to face this problem in a fast and effective way [1]. In such a framework, learning-by-examples (LBE) techniques [2][3] have demonstrated to be promising solutions able to enable detection procedures efficient in terms of both resolution and required time/computational resources. This paper is aimed at describing the detection problem as a three-dimensional classification process and analyzing its extension from theory to real experiments through a careful numerical analysis. Thanks to an integrated strategy based on a Support Vector Machine (SVM) classifier and a multi-resolution approach, a multi-resolution detection is obtained by means of an iterative zooming that considers only the regions characterized by an high probability to be occupied by the buried object. The arising time and computational saving allows the definition of an high-resolution map despite the complexity of the three-dimensional scenario at hand.