We aim to extend current image-based modelling technologies such that they may be applied to historical images, which were taken for other purposes using film cameras that have long since disappeared. This will require the development of new self-calibration techniques applicable to analogue images, and interactive modelling mechanisms capable of extracting 3D from unconventional image sets. In the most extreme case, there is only single image available of an object to be modelled, but in general there may be multiple images, taken over a long time period. The appearance of the object to be modelled may have changed significantly between images, due to lighting and weather variations, but more substantially, due to changes in geometry. Figure 1, for example, shows 3 images of the HMAS Protector. These images were taken over a period of tens of years, during which time many additions were made to the shape of the ship. The locations of the images vary significantly also. Little of the scene geometry thus remains constant over the image set, and yet humans are able to estimate the shape of the ship and its surroundings. By exploiting the human capacity to recover shape estimates under challenging conditions we aim to provide 3D modelling software wight he same capacity.
This project aims to give historians the capability to develop their own 3D models, directly from historical photographs. This will be achieved by extending existing structure-from-motion techniques to meet the demands placed upon them by the nature of the images. The interactive modelling process will be such that the models developed will be perfectly suited for incorporation into the digital exhibition tools currently used within museums, and more broadly.
This work is funded by ARC Linkage Project LP130101064 held by Prof Ian Reid and myself which has as its Partner Organisation the Maritime Museum of South Australia.