Range-based versus automated markerless image-based techniques for rock art documentation

Abstract

Nowadays there is a huge proliferation of fully automatic image-based solutions producing either three-dimensional (3D) point clouds or 3D models. However, the reliability of the output is not usually reported and clarified. This paper presents a comparison of the 3D modelling results achieved on two rock art shelters at separate archaeological sites using a high-resolution digital camera. The 3D point clouds were produced using automatic image-based photogrammetric and computer vision software running either locally (FOTOGIFLE and VisualSFM) or through a webbased reconstruction service (Autodesk 123D Catch). The first two automatic approaches are compared with a manual bundle block adjustment. Dense image matching was undertaken to densify the point clouds prior to 3D modelling. The derived 3D models are compared with a time-of-flight laser scanning dataset that was used as ground truth. Statistical analysis shows significant metric differences on the digital surface models for the different solutions.