Unmanned Aerial Vehicles (UAV) and photogrammetry are two fields that have been boosted these last years. Using aerial means, one can easily acquire aerial data and produce high resolution dense surface models, orthophotos,... <br><br> IGN (the French Mapping Agency) and CNR (Compagnie Nationale du Rhône, which is the concessionary of the Rhône river and a hydraulic energy producer) have associated themselves on a thesis protect. The aim is to be able to monitor dykes from images acquired by UAV and take benefit from their convenience, targeting a centimetric accuracy on the Z-axis. This article presents our motivations and the problems we have faced in our first experiments. We also worked on a site covered by a terrestrial Lidar survey, and studied how minimizing the bundle adjustment residuals by using different calibrations would influence the quality of the computed models. Finally, we will introduce in a last part our last experiments to get a better understanding of poses estimation accuracy.