Future European missions aspire to bring soil samples from Mars back to Earth. The precise identification of the site from which samples are taken leads to major constraints as to the landing accuracy of the lander. Vision-based absolute navigation now appears to be the most promising navigation method for correcting errors originated from the interplanetary voyage at the point of re-entry. The thesis has consisted in improving the vision-based absolute navigation simulator developed for the landing phase. The image processing and identification algorithms have been based on algorithms implemented in 2012: matching between the characteristic points of the embedded image and the points which come from processing the measured image. The work has been focused on improving the navigation method with the main objective of minimizing the onboard design load and improving the robustness of the navigation system performance. The main tasks to continue the development of the simulator have followed the next steps: - Implementing and validating an iterative algorithm supplying a relative position/attitude in relation to the landing site. - Implementing flexible hybridization, incorporating the level of confidence in the matching carried out. - Studying the problem of the navigation filter initialization for a Martian landing.