This thesis presents the development and verification of concepts, methods and possible implementations of a system for the automatic visual inspection of deep mine shafts. Although the concept is developed for shafts, it can be applied to any large object for which a contiguous visual surface inspection is required. The concept is based on mosaicking large numbers of pictures, e.g. 60000, to structured panoramas which can be mapped as texture on a geometric model. The large number of images together with their resolution enable a precise metric representation of the observed surfaces. These are stored in a reduced resolution set format. This data model is consistent with open source mapping tools so that data fusion from different sources, e.g. depth based geological information, may be performed. Standard web based visualisation tools can then be used to view the data.