Feature based ultra-wideband object recognition

For the goal of an Object Recognition (OR) in emergency situations, an OR Ultra-Wideband (UWB) Radar system is proposed in this paper. Conventional OR Radar systems based on vector machines or neural networks result in a high recognition rates, but are not suitable for OR in a real time scenario, due to the vast computational load. Hence the OR Radar system proposed in this paper is based on a minimum mean square error detector and seven Object Recognition features with low mathematical and computational complexity. Furthermore, the proposed OR features are extracted from polarimetric images Radar acquired by two imaging methods. Experimental validations are performed with an alphabet of twelve complex objects, a M-sequence UWB Radar device (4.5 GHz - 13.5 GHz) and compact dual-polarized Ultra-Wideband antennas.