While we have been using Pix4D for aerial surveys, we have adapted the technique to perform surveys of coral reefs. For obvious reasons, GPS referencing is a challenge underwater. So we include in our survey a metric rod as a dimensional reference and that supplies us with fairly accurate measurements.
But we do experience annoying problems with the general orientation of the models. Our technique consists of performing 3 passes over a 10x2 meters transect i.e. one pass at the apex and 2 passes at 45 degrees on either side of the transect. The key issue is that Pix4D has a hard time figuring out the xyz orientation and, as a result, the mosaics are unusable. The problem stems from the fact that the reef has a very irregular surface with little or no horizontal plane reference. In an effort to correct the problem, we spend hours trying to orient the model with orientation constraints but here again the very rugged nature of the substrate makes that task very challenging. We tried using the metric rod as a plane reference but that fails 80% of the time since it is very difficult to insure that the rod is lying on a flat surface.
Since the mosaic is built from what Pix4D thinks is the apex, the mosaic is often rendered at an angle that makes the results unusable given that the perspective angle distorts the images used to render the mosaic.
I have two potential solutions in mind. The first one would be to let the user orient the dense cloud visually by locating the point of view at the center and apex of the model, and rotating the model in the xy plane… That is fairly easy to do since the shapes will visually align along the vertical line of sight and implicitly define an acceptable xyz orientation. A button allowing the user to set the orientation from the view would be very helpful.
The other solution would be to allow the user to define the point of view from which the mosaic should be rendered. The approach of an arbitrary rendition angle might actually be useful to other people.
I do realize that both of these methods may somewhat compromise the absolute metrics of the model, but they would at least provide a way to get an acceptable mosaic.