3D seismic reflection is widely used in the oil industry. This standard subsoil auscultation method provides information on geological structures and can be used to build reservoir models. However, the properties derived from3D (and 2D) seismic data are only static: 3D does not allow to evaluate the changes with calendar time. The addition of a temporal dimension to 3D data is obtained by repeating the measurements at several dates separated by several months or even several years. Thus, 4D seismic (time-lapse) makes it possible to measure and to analyze the changes of the subsoil in the long term. Since the 90s, this method is used worldwide at sea and on land. To carry out a much more frequent monitoring (daily), even continuous (a few hours) of the subsoil, CGG developed, in collaboration with Gazde France (now ENGIE) and Institut Français du Pétrole (now IFPEN), a solution based on buried sources and receptors: SeisMovie. SeisMovie was originally designed to monitor and map the gas front in real time during geological disposal operations. It is also used to observe the steam injection required for heavy oil production. In this thesis, we bring contributions to three challenges arising in the processing of seismic data from this system. The first one concerns the attenuation of near-surface variations caused by "ghost" waves that interfere with primary waves. The second one concerns the quantification of subsurface changes in terms of propagation velocity variation and acoustic impedance.The third one concerns real-time: the data processing must be at least as fast as the acquisition cycle (a few hours). Infact, the analysis of the data must enable the reservoir engineers to make quick decisions (stop of the injection, decreaseof the production). In a more general context, there are conceptual similarities between 3D and 4D. In 4D, the repeated acquisitions are compared with each other (or with a reference). In 3D, during acquisition, field geophysicists compare unitary shot points with each other to assess the quality of the data for decision-making (reshooting, skipping orcontinuing). Therefore, some 4D real-time tools developed during this thesis can be applied. A new approach called TeraMig for automated quality control in the field will also be presented.