The content of this thesis is focused on the study of an UWB pulse radar for through the wall vision which has many applications in the military domain (assaults, hostage rescue,…) and in the civil security domain (people search in the rubble, in a fire).For these uses, the real observed scene image is not necessary, only some relevant information is enough: number of persons, position, velocity of movement, etc. It’s in this context where we have developed an experimental detection device and some through the wall targets localization techniques. The developed radar consists of a pulse transmitter covering the frequency range from 3GHz to 6GHz, and three independent receivers, combined with some localization algorithms and image reconstruction.The first algorithm is based on a technique of trilateration. Once the theoretical model is presented, many resolutions methods are studied for estimation of the signal propagation distance measurement. The Brent-Dekker method has been chosen for its fast convergence and low number of iterations. Although the trilateration technique is accurate, it does not allow obtaining a scene image, but just locate the targets, without providing their dimensions. Therefore a second approach, based on a “beam forming”, has been studied, in order to obtain a 2D scene representation. We have chosen to develop a non-coherent backprojection method, it is the most simple technique which does not require a strong constraint on the signal phase nor on the antenna positions. Different backprojection methods have been proposed: the backprojection with cross correlation, the improved backprojection with cross correlation and bi-cross correlated backprojection. These three methods are based on a fusion of captured information by the receiving antennas and with the obtained information on the “referenced” antennas. This fusion allows improving the image quality progressively.To conclude this work, a perspective is initiated in order to minimize the false detection rate, it is based on the cooperation of the trilateration technique and bi-cross correlated backprojection. This cooperation allows matching the estimated detection by the trilateration technique and the bi-cross correlated backprojection technique. The mismatched targets are removed according to some specific criterias to each approach.
