Supercomputing at the trench edge: Expediting image based 3D recoding
David Stott, Matteo Pilati, Carsten Meinertz Risager, Peter Jensen, Casper Skaaning Andersen, Jens-Bjørn Riis Andresen
Abstract Image-based 3D reconstruction with Structure From Motion (SFM) techniques are increasingly used for documenting archaeological excavations. They afford an inexpensive means of recording accurate, detailed spatial and radiometric data, but there are a number of challenges presented by these methods. Foremost among these is the time it takes to produce a finished model. Due to the computationally intensive nature of these techniques processing often takes hours to complete. This is problematic, as archaeological excavation is by its nature a destructive practice where success is contingent on the quality of the record. Validating and interpreting these data in the field, while the subject still exists is essential. Thus, rapidly processing the models and making them available to archaeologists as they excavate is of crucial importance.
In this presentation we demonstrate how the process of model generation can be expedited to occur in near real-time, from the field. This is achieved using three approaches. First, by remotely processing the images over a 4G mobile internet connection to the High Performance Computing (HPC) cluster at DeIC Abacus 2.0 we can dramatically increase the computational power available on site. Second, we examine the effect different parameters have on processing speed and quality of the finished model. Third, we argue that data transfer and processing time can be further optimised by appropriately constraining image size to the scale of the objects being recorded.
