Open source software for numerical computation

Skip to main content

Automatic detection and characteristics extraction of archaeological structures using coupled Geographic Information System software and Scilab

Jean-Pierre Toumazet, GEOLAB

Since the last few years, archaeologists have the opportunity to use a very powerful tool in order to detect archaeological structures located in forest terrain: the airborne LiDAR. A Laser beam, emitted from an airplane, partially goes through the vegetation and is reflected on the ground. A measure of the time necessary for the light signal to return to the source allows then, knowing precisely the position of the transmitter, to reconstruct the relief very accurately. 
The cost of this tool tends to decrease, and it is now used in almost all the research programs dealing with archaeology when at least a part of the studied area is located under the vegetation.
Resorting to the LiDAR yields a huge quantity of data, constituted of clouds made of millions of points from which it is necessary to extract the relevant information. A manual processing of these data is then boring, time consuming, and even sometimes impossible: the automation of the detection process becomes necessary.

The presented work deals with the process of automatic detection of archaeological structures. It is applied to former agricultural constructions, built from the medieval to the modern period. They can be found in very high densities in some places in Auvergne. These structures have been chosen to test the process of automatic detection because they are particularly delicate to treat: they are indeed very variable in forms, appearing sometimes isolated, sometimes in group and they are characterized by small relief variations very difficult to detect among the global relief.

The process presented uses the associated potentiality of Scilab and a Geographic Information System software. At first, the point cloud is interpolated in order to obtain a Digital Terrain Model (DTM). This DTM is then smoothed, by a moving average filter, in order to erase the abrupt relief variations, generally from anthropogenic origin: the macro relief is then highlighted. It’s then removed from the initial data, so that only the micro relief becomes visible. We thus obtain a Local Relief Model (LRM) in 2D ½.. An algorithm of automatic recognition allows then to detect the structures, to classify them, to extract their geometrical characteristics and to create a database in a totally automated way.

Paper signed by Jean-Pierre Toumazet, Erwan Roussel, Marta Flores, GEOLAB, CNRS, Bertrand Dousteyssier, Franck Vautier, Université Blaise Pascal, CNRS.