I’ve been using RealityCapture from time to time. And my biggest beef is that if I record photos of a single object, that RealityCapture creates several components. And those components are really hard to fuse, to make into one model. So this is my suggestion for a new workflow. All models should be layered on top of each other. And on the left panel you should be able to select one and rotate it in the right position. Then you select the next level and rotate it so it fits roughly in the position where it should be. And you continue doing this. You put the point clouds where they should be. And then you have the option for RealityCapture to align based on your pre-alignment. So it knows where’s the connection. And then it finds the points. And then it finds the connection. Then it creates a good model. Because so far with finding the same point in pictures, and it still doesn’t work, it would be so much better. Because I often see myself think that “can’t you see that? Can’t you see that this model is here and this model is there? And this is how they are connected. Can’t you see that?” And giving it this hint by aligning them. Or maybe having a few points to align. And also talking about alignment for plane alignment. Why don’t you just give the opportunity to select three points on a plane of the point cloud model. And then it aligns the whole model based on that on the ground in the center of the system. Or you say “alright this point is similar” or “it’s the same like this point and this point”. And you do this again with three points. And then it’s in tune. But it never seems to really work. And you are working with images, you are working, you are confusing yourself. So this would be so much easier.
Thank you @Kurt_Sennerich for your feedback, I added it to our database. For now, the best way to prevent multiple components and the need to use any control points is to take good quality photos with 100% coverage of the scanned object and with information overlap between photos to achieve successful, high quality reconstructions. This is valid for any scenario from a turntable scan of a small object, building scan, or a large scale mapping project. We have a lot of resources about the propper scanning process in our learning library: