Estrazione di punti omologhi per l'orientamento automatico di sequenze di immagini

The geometric characteristics of a building represent a topic of considerable importance for cadastral purposes. The reconstruction of building 3D models can be obtained trough laser and photogrammetric surveys, with the possibility of measuring the volume and to produce a 3D cadastral database to be shared inside of the public administration. Such surveys are currently implemented either by aerial or terrestrial platforms, with cost not negligible for the public administration. Low-cost methodologies are nowadays being developed to be accessible and easily usable by local authorities with limited budgets. The theme in this work is part of a research project intended to identify low-cost and automated methods for the recognition of buildings in order to determine their geometric features. Such project includes the use of photogrammetry by amateur camera or video camera mounted on radio controlled mobile vehicles or that follow a predestined trajectory. The amount of data obtained with this acquisition method can be massive, therefore it becomes essential the automation of some photogrammetric process phases. In this work we expose results obtained from a specific part of the project dedicated to the creation of a procedure for automatic selection of homologous points on multiple images to the orientation of the photogrammetric block. The algorithm used is the SIFT (Scale Invariant Features Transform), a valid alternative to classic feature extraction algorithms used in aerial photogrammetry, particularly useful if the photogrammetric images are taken in conditions far from a normal geometry. The procedure provides for the possibility to abate the time for calculation and filtering of possible false matches arising from the matching procedure through the application of robust methods. Some results are reported relatively to the production of digital models for the surveyed artefacts obtained through direct communication modules of feature extraction achieved in MATLAB with those of Bundle Adjustment and image matching in the commercial software PhotoModeler Scanner