Multi-image Matching

In close-range architectural and archaeological photogrammetric applications, very often, amateur cameras are used. While images are suffering from excessive distortions, the objects themselves exhibit remarkable anaglyph (walls, excavation holes, 3D objects, etc.), prohibiting thus the use of simple planar rectification. The 3D reconstruction of such objects unavoidably requires the production of 3D Digital Surface Models and the subsequent production of orthoimages. But then, the main problems we are faced with, are the big number of images, their arbitrary convergence, the need for automation in object reconstruction, in often cases repeated patterns (eg. stones, bricks, soil), unfavorably low image texture, etc. All these factors create a very unfavorable situation for the application of automatic image matching techniques. Moreover, the matching algorithms in commercial photogrammetric systems are mainly referring to the aerial case, where most of the above problems do not exist. For obvious reasons, also, they do not provide for multiple image matching. This paper deals with the development of a flexible multi-image matching technique under the above mentioned situations. Its novelty is placed in the development of the mathematical models used but mainly on its ability to cope with harsh situations, often met in architectural and archaeological applications. In order to show the algorithm’s performance extreme but real examples are shown, where subpixel accuracy is achieved.