This section will give you an short overview on how the documentation of objects with Structure-From-Motion works and where are the limits of this technique.
The methods are well known in geodesy and used for triangulation of surfaces using photos of (manned) airplanes. With increasing computation power of personal computers and progress in many algorithms it became possible to adopt and improve this approaches for different research projects, especially in computer vision. One of the first projects using large image collections and the approach of Structure-From-Motion was the “Building Rome in a Day” project at the University of Washington. From then more and more algorithms and software suits were developed, leading to the now available commercial and non-commercial variety.
Different SFM softwares are available, commercial and non. Very widely used by Archaeologists is Agisoft’s PhotoScan, providing a all-in-one software with quite good value for money. For the use with drones there is Pix4D, showing very nice results, especially for mapping purposes.
In the non-commercial sector is VisualSFM by Changchang Wu the most important software. It is easy to use and has – with some download effort – a complete workflow from pictures to model. This is the program I use for the camera position reconstruction. But instead of using it’s PMVS/CMVS tools for a dense pointcloud reconstruction I am glad to be allowed to use NFrames Sure. This is a versatile and highly precise software, developed at the Institute for Photogrammetry at Stuttgart University. It is really easy to use and delivers pointcloud of very good quality. Some of them will be shown at the results section in the Structure-From-Motion section of this website.
This method has limits – of course. It struggles heavily with objects of homogeneous colour or low contrast surfaces. Also reflections and moving shadows are a problem while working out in the field.