Abstract - This paper focuses on an emergency situation in which a network of ultra-wide-band (UWB) sensor nodes mounted on moveable platforms is moving in a room for purposes of recognition of objects. The recognition is accomplished between 2D canonical objects, which are scanned by UWB Radar and a set of reference objects with same size which were analyzed by a ray-tracing method. The object recognition task is performed on the basis of RCS (radar cross section) measurements leading to corresponding radargram data. Here the objects characteristic geometric features are examined by using polar Fourier descriptors which are extended in a way that also provides for multiple reflections. Keywords: UWB radar, Object recognition, polar Fourier descriptors, UWB radio technology 1. Introduction Radars utilizing UWB pulses have an advantage of high-resolution measurements of time-of-flight data or distance respectively. Due to the high-resolution it is quite insensitive against multi-path conditions. They can also be used in emergency situations where dark smoke and dust disallow optical measurements. Therefore, UWB pulse radar is attractive as an environment measurement method for various applications including moveable security-robots. To be able to perform these tasks the sensors have to localize themselves and all objects in the room, and then perform an inspection for refining the object images and recognize them. This paper deals with the subsequent recognition of the objects. Based on the previous localization and imaging the position and the rough form of all objects in the room are known so that possible tracks for a second inspection tour of an additional sensor node around objects can be determined. This inspection tour provides data for image refinement and object recognition. To achieve that the surface of each object is scanned step-by-step using UWB antennas with very small opening angles. We have already proposed a recognition algorithm based on the moment invariant method [5] in [1] where the peak data or the reflections of 1 st order respectively is utilized. To enhance the robustness of the recognition algorithm and to enable a better performance with complex objects multiple reflections due to detailed structure of the object shape must be evaluated within the recognition algorithm. First and foremost, the objects geometric structure and characteristic has to be extracted as accurate as possible. To resolve this problem, in this paper we propose a robust recognition algorithm based on polar Fourier descriptors. These Fourier descriptors are extracted both from the reference and measured radargram and are compared with a least-error classifier to the power of 3. The reference objects are investigated with a ray tracing algorithm providing the simulation based reference radargrams. 2. Ray Tracing based Reference Data We assume LOS conditions to enable a robot motion and for simplicity we assume a full reflective surface and do not evaluate the interiors of the objects. The used objects consist of simple canonical and some non-canonical complex objects in the form of beams with no variance in the 3 rd dimension. The following 12 Objects were used: Figure 1: Contour of the used 12 Objects Robust UWB Radar Object Recognition R. Salman 1 , T. Schultze 1 , M. Janson 2 , W.Wiesbeck 2 and I.Willms 1 1 Fachgebiet Nachrichtentechnische Systeme, Universität Duisburg-Essen, 47057 Duisburg, Germany 2 Institut für Höchstfrequenztechnik und Elektronik, Universität Karlsruhe (TH), 76131 Karlsruhe, Germany E-Mail: salman@nts.uni-due.de, malgorzata.janson@ihe.uka.de