Human Development in Landscapes | People | Doctoral research student Robert Wulff, Diploma

Image-Based 3D Documentation in Archaeological Trenches.

Documentation plays an important role in archaeology and particularly the documentation of trenches requires much time and effort during an excavation. Common techniques for documenting archaeological trenches include photographs, rectified photographs and vector-based CAD plans. All these methods have in common that they produce 2D representations.

This project investigates the application of methods from computer vision and computer graphics in the research field of archaeology. The goal is to aid archaeologists in their work, specifically by focusing on the 3D documentation of archaeological trenches. Producing 3D models of such trenches is achieved by performing an image-based 3D scene reconstruction (“Structure-from-Motion”). Though the accuracy of such methods is usually lower than accuracy when using laser scanners, they only require equipment that is already part of the archaeological documentation workflow, namely a digital camera and a total station. Structure-from-Motion allows the reconstruction of 3D models in an arbitrary coordinate system only. The total station is therefore needed to transform the models into the reference coordinate system used at the excavation site.

Another key aspect of this research project lies in the post-processing of the data. A 3D model only provides geometric information about the reconstructed trench. In contrast, a CAD plan is an abstract representation: it “encodes” the semantic entities (finds and features) found in the trench. One goal is now to combine these two domains by performing a semantic classification on the 3D models according to the associated CAD plans. Thus, it will be known for every part of the 3D model to which archaeological entity it belongs.

In addition to enriching the models with semantic information, temporal aspects will also be considered. An excavation is not static, but a 3D model is just a snapshot of a trench's configuration at a certain time step. It is therefore desirable to reconstruct 3D models of all layers of a trench and to align them automatically. This may also allow automatic temporal ordering, because the trench boundaries (“profiles”) only grow in size throughout the excavation.

Having determined the semantic entities of a trench opens a new perspective concerning data integration: It allows all photographs that are registered against a model to be labeled automatically according to the semantic entities that are visible in it, i.e. for a certain image it becomes possible to automatically determine which part of it belongs to which archaeological entity. So these techniques not only allow classifying parts of 3D models semantically, but also images. This will lay the foundation for comprehensive data integration from all the different sources. A database might handle this to allow users, for example, to query for all available representations (photographs, drawings and 3D models) of a certain find.