University of Southampton OCS (beta), CAA 2012

Font Size: 
Image-Based 3D Documentation of Archaeological Trenches Considering Spatial, Temporal and Semantic Aspects
Robert Wulff, Reinhard Koch

Last modified: 2011-12-17


Image-based Computer Vision techniques have become increasingly popular for documenting archaeological trenches. Such techniques provide a means for reconstructing the 3D geometry using only equipment that is already part of the archaeological documentation workflow, namely a digital camera, a computer and a total station for transforming the 3D model into the reference coordinate system used at the excavation site.

This paper presents a scene reconstruction pipeline suitable for image-based 3D documentation of archaeological trenches. However, the main contribution lies in the post-processing of the data which enriches the geometry (= the spatial aspects) with temporal and semantic information.


In the post-processing step the model is first classified coarsly to locate the trench boundaries (profiles) by clustering the surface triangles according to their orientation and position. The profiles are usually perpendicular to each other and can therefore be found by searching the clusters.

Next, the models of different excavation layers from the same trench are registered against each other by establishing keypoint correspondences between them. These keypoints are restricted to the profiles located in the previous step as these only grow in size and remain relatively static throughout the excavation. The observation that the profiles only grow in size can be exploited to perform an automatic temporal ordering of the single models. This incorporates temporal aspects and allows to reproduce the single steps of the excavation virtually.

Since the models of the single layers are registered against each other, it is sufficient to transform only one of them to the reference coordinate system from the excavation site using 3D points aquired with a total station. The transformation can then be applied to all remaining models accordingly. This allows measuring in the models and correlating them with other spatial data.

Finally, the geometry is augmented with semantic information by classifying it into archaeological entities (finds and features) in addition to the profiles that have been located in the first post-processing step. The basis for this classification are CAD plans which are produced directly on-site using a total station. As these 2D drawings reside in the site's coordinate system, they can be related to the 3D geometry of the trench. Since the drawing has a lower dimensionality as the model, problems arising at perpendicular surfaces and protruding objects need to be handled explicitely.


The innovation of this paper lies in the holistic approach of combining spatial, temporal and semantic aspects of archaeological trenches into a unified frame. Augmenting the geometry with semantic and temporal information allows archaeologists to carry out the interpretation of the finds and features directly in 3D space.


scene reconstruction; 3D documentation; semantic classification; segmentation; virtual archaeology