This thesis is devoted to object localization, using high resolution methods, in a large band and multipath background. From array processing algorithms adapted to time of arrival estimation, we propose a new method for the localization of real (broadcasting) or virtual (reflective) sources of a known, large band signal. The originality lays in the incoherent way of treating data. On each ensor, time of arrival are estimated. Then, they are linked to a source, enabling its localization. However, this methods suffers from the signals coherence. That is why we propose a frequential smoothing preprocessing that helps lowering this correlation while maintaining an optimum resolution. The performances of the whole method are shown on both simulated and experimental data.
