Georeferencing of point clouds in VisualSFM

0
540

One of the greatest concerns when dealing with point clouds is the need to georeference them (transformation based on Ground Control Points), or to contextualize them within a coordinate system.

Fortunately, in the latest versions of VisualSFM, the author has incorporated the option “GCP-Based Transform” (Ground Control Points) located in the sfm > More Functions menu, to assign coordinates in an XYZ system or also the GPS-based Transform option, to assign geographic coordinates.

With these options, we are offered the possibility to select a point in our cloud and with CTRL + Left Click, fix the point, after which the corresponding real-world coordinates for that point will be requested. Once the process is finished, the command SHIFT + Menu Click must be executed, where Menu is the path of the command sfm > More Functions > GCP-Based Transform. After the process, the dense cloud creation must be launched again, or the project saved (Save NV).

The truth is that there are not many free options to carry out this process, which tools like Photomodeler or Agisoft Photoscan perform without many problems. When creating point clouds from SfM techniques, we find that the resulting models are not contextualized in space, making it very difficult to relate various models to each other. There are processes such as Bernhard Fritsch where the GIS tool GRASS 7 is used through the v.ply.rectify plugin and with the help of editors like Cloud Compare, allowing us to associate a list of real coordinates to specific points in our cloud, enabling easy relation of several models in tools like CloudCompare or MeshLab, as well as performing measurements.

The difficulty we face is identifying a point with some precision within the discrete cloud. Therefore, often the simplest approach will be to generate the dense cloud to locate our GCPs with some ease. However, the spectacular feature of VisualSFM is that you can also assign GCPs on the raster images of your project, so that the localization of targets or points of interest can be done with very high precision. In general, as we switch views to sparse cloud, dense cloud, or raster, if we have the GCP-Based Transform tool applied, we will be able to assign control points.