Image Combination into Large Virtual Images for Fast 3d Modelling of Archaeological Sites

In archaeological sites there are often several levels of ruins when the site has been constructed upon other ancient or prehistorical sites. It is very important for the archaeological research to map and monitor each step of the excavation. The traditional way for mapping of excavation is time consuming. Digital photogrammetry enables instead fast acquisition and processing of the data. Continuous excavation reveals different levels of findings coming from different chronological periods, images have to be acquired before findings from each level are removed and excavation continues. Therefore, the storage of data and processing time increases due to the number of images acquired from the excavation site. Apart from that, the images are acquired from amateurs, with less experience in acquisition techniques. A method has therefore been developed to transform many individual images into larger virtual images which can be considered of being of a normal central projection. Images can be combined under the assumption that the accuracy is sufficient for archaeological documentation. For the generation of each undistorted virtual image the Brown distortion model has been used. The model gives the corrections from the distorted to the undistorted image, therefore the invert transformation has to be calculated through an iterative approach, using the non-linear set of Brown equations. For the generation of the virtual images 2 strips were used, each consisting of 3-4 images with 70-80% overlap. The strips form two virtual images with sufficient overlap to extract a DSM. The images had to be enhanced and be radiometrically balanced. In this way more features can be extracted and mapped, even small structures that are covered with dust or lie in shadowed or overexposed areas. Methods are presented to reduce noise, enhance edges and contrast. They are essential for higher image quality and further processing. Results from automatic DSM generation using virtual images are presented, using a new multi-pass adaptive matching algorithm. The algorithm uses features as primitives. Through a multi-pass technique at each extracted feature and computed quality measures, a more robust quality control ensures higher reliability of the final result. A comparison of automatic DSM generation from the separate models of raw images and the two virtual images is done. All results are compared with a manual extracted DSM and presented.