BENCHMARK OF METRIC QUALITY ASSESSMENT IN PHOTOGRAMMETRIC RECONSTRUCTION FOR HISTORICAL FILM FOOTAGE

Abstract. Quality assessment in photogrammetric processing is fundamental to obtain metric information and to reconstruct 3D models of Cultural Heritage, especially when it has been lost or changed over time. The determination of metric precision is technically challenging when dealing with historical films and videos that in many cases represent the only remaining traces of this heritage, which is useful for architectural, archaeological and restoration studies. This paper examines the suitability of existing photogrammetric software to evaluate the maximum possible metric accuracy for processing videos shot with fixed camera motions. In order to evaluate the metric quality obtained processing historical film footage with photogrammetric techniques, a benchmark was created on a new video dataset with the aim of reproducing the camera motions in which old video were shot. Three different camera motions were considered: Up/Down Motion-Tilting, Left/Right Motion-Trucking and Rolling Motion-Panning. The methodology was experimented on Valentino Castle in Turin, a monument inscribed in the UNESCO World Heritage List. Data were processed with the implementation of open source Structure-from-Motion algorithms and the results were analysed for the evaluation of metric quality. Results show the different maximum precision assessments according to the different typologies of camera motion. This research provides fundamental support to historical studies on Cultural Heritage, creating a sharing standard with zero-cost data and tools useful for both geomatics and restorers.