Data on the objects of the environment are, in general, entered in a computer database where they are indexed by their 3D coordinates and some descriptive features as the length, width, orientation, number of faces for the tools, etc. Nevertheless, the shape of the objects is so complex that a representation by a small number of parameters is insufficient. Moreover, the indexing is performed manually and the description can be biased by subjectivity as the excavation team changes every year, or by tiredness as several thousands objects are found! A digitalization of the shape and of the texture would allow building a 3D digital model which would be a complete representation of the object. This could also lead to an automatic extraction of complex features to improve the indexing and classification.
The first step will be to make a state-of-the-art of the tools to acquire environment data. They have to be fast as a lot of items are found during an excavation campaign, accurate to emphasize the small details that can be of high importance for paleo-anthropological studies as the facets of a stone tool, portable and solid in order to be carried on the excavation site and, if possible, cheap! In particular, we will study the tools based on the reconstruction of the shape from a set of video-images. By using standard digital cameras, it is possible to obtain simplified 3D models at low cost but we have to test if they are accurate enough for paleo-anthropological research. More precise photogrammetric techniques exist but they are more expensive, slower or require static equipment. An alternative is to use laser range finding devices that give very precise models but they are quite expensive. In both cases, we obtain also the "texture" of the objects that can be very informative as in artistic objects.
Indexing the objects in the excavation database is currently done by hand and is, in general, based on very simple features (as the object type and some orientations or maximal lengths). A more precise analysis can be performed by a typology expert but it takes time and can only be performed after the excavation campaign where several tens of thousand objects are found! Moreover such procedure is subjective and can lead to misinterpretations in the database if all the experts do not agree on common criteria. The second and third step of our research will consist in studying the very topical problem of indexing automatically 3D models. The process could be based on some 3D geometric features that are automatically extracted as salient lines or the flat parts. For example, such landmarks could emphasize the facets of a stone tool and help to categorize it as a biface or a chopper. An other idea for classification is to compute a "shape distance" by registering the 3D model with several reference models and quantifying the deformations. Then, the "closest" reference model gives the object category.
The Prehistoric Man cut the stones according to their final utility but also to his cultural behaviors. Flakes were successively separated from the core stone and left on the excavation site. By identifying sequences of conjoining lithic artifacts (flakes and the core) and fitting them together, it becomes possible to recover the cutting process and understand the stone tool manufacturing activities. Refitting lithic artifacts by manual trial and error is an accurate but labor intensive method. Automated 3D surface matching of conjoinable artifacts will enhance the efficiency of this valuable analytic method. We plan to study this very innovative topic in step 4, by developing some matching methods based on 3D geometric landmarks.