Par le principe du temps vol, un système émettant un pulse peut mesurer à quelle distance se trouve des cibles en calculant les délais d'arrivée des échos retournés par les obstacles. Des situations de détection complexes doivent être résolues, telles que deux cibles dont les échos se superposent partiellement. L'algorithme de détection développé a donc pour objectif de distinguer des cibles rapprochées entre elles, de façon fiable et précise, avec une bonne immunité au bruit, sur des signaux exclusivement réels, où seule l'information sur l'amplitude de l'enveloppe du signal est disponible. L'algorithme développé sera basé sur l'algorithme MUSIC. Ce dernier est inutilisable tel quel, dû à la nature des signaux. Une adaptation est tout d'abord élaborée, puis optimisée. Cette version de l'algorithme surpasse les performances des algorithmes de notre partenaire industriel et des méthodes de détection généralement employées et est en mesure de distinguer les échos de cibles rapprochées entre elles. / By using the time-of-flight principle, a system emitting a pulse is able to measure the distance of a target by calculating the echoes' delays returned by the obstacles. Some complex detection situations must be solved, such as two targets producing overlapping echoes. The goal of the detection algorithm is to distinguish targets with overlapping echoes, with a good precision and a good immunity to noise, using real signals, which only the enveloppe's amplitude information is available. The created algorithm is based on the MUSIC algorithm. The later is not working as it is, because of the signals' properties. An adaptation is created, and then optimized. The most substantial improvement comes from the decorrelation processing applied on the signals' covariance matrix. The effect is a decorrelation of the sources, allowing the algorithm to distinguish targets with overlapping echoes. Also, most of the decorrelation techniques help to detect echoes with low SNRs. Another improvement concerns the measurement resolution, which is better than just the sample period. The algorithm's performances exceed those of our industrial partner algorithms and those of commonly used detection methods. The ultimate goal of the project is to integrate the developped algorithm into our industrial partner's system. It has to be real time application, and to respect the cost and ressources constaints of the system. Consequently, some optimizations of the algorithm were required. Some specific properties of the covariance matrix allowed a decrease of the memory space to save its data. This way, the number of matrix's data saved represents less than 5% of the initial covariance matrix. Another optimization is done by using an iterative method for the eigenvalue decomposition, accelerating significantly the processing time. Finally, the algorithm 'sperformances coming out of the comparative tests completed between the adapted MUSIC algorithm and our industrial partner's algorithms demonstrate that the project's goals are fullfilled. The developped algorithm can solve the situation where two targets produce overlapping echoes, while providing a good noise immunity.