Marco BLOCK-BERLITZ1 | Raul ROJAS2
(1HTW Dresden, University of Applied Sciences, Dresden, Germany | 2Free University Berlin, Germany)
Keywords: UAV, 3D videogrammetry, best practice, keyframes
The videogrammetric approach, i.e. the use of videos or continuous single camera shots with high frame rates for 3D reconstruction, was introduced ca. 15 years ago. It represents a practical, less risky alternative to ambitiously planned single shots and seems to be more cost-effective, faster and easier-to-use in many cases. It is also more robust against partially defective data. The high frame rates allow continuous movement while recording the data. Therefore it is suitable for use with UAVs and many other mobile systems. But the biggest advantage of the video-based compared to the photo-based approach is the amount of recorded data. If the initial reconstruction process (with a reduced subset of video frames) produces holes or otherwise incomplete models, more images (frames) can be added to the process in order to iteratively refine the model. In contrast to a collection of single images, more mathematical correlations can be extracted from video streams.
The videogrammetric approach is a promising candidate for producing low-resolution, initial 3D models on-site, in very short time. If the data is adequate, it is possible to at least ensure connected, gapless 3D model reconstructions. The more time-consuming maximization of model detail can then be delayed to an off-site post-processing stage.
We present a videogrammetric workflow and discuss best practice for documenting excavation areas with low-cost UAVs and action cameras. We also discuss an efficient, automated approach to extracting suitable keyframes from video streams. The software JKeyframer was developed to fill the gap between video data recording and SfM and is freely available. It analyzes video streams and automatically extracts suitable sets of keyframes. We show results of a wide range of synthetic and real data experiments showing that videogrammetry is a fully viable approach to reconstructing single objects or complete archaeological areas.