![]() Before I move onto the next stage, I’m going to crop off a little more from my models – the ‘down’ mesh (purple in the image above) has a little too much on the top that isn’t well reconstructed: I’ve just deleted a bit of ‘down’ (purple) from the top. I’ve removed the textures so you can see the meshes more clearly: When you click on ‘Align’ the meshes will move so that they minimise the cumulative distances between all the points: Roughly aligned models Because photogrammetry is inherently scale-less, there’s no guarantee the models will be the same scale, so you need to let cloud compare change the size, not just the translation and rotation. More wouldn’t hurt.īefore you click ‘align’, make sure ‘adjust scale’ is ticked. I’ve selected 8 corresponding points here. I think 3 or 4 is the minimum, but you should always aim for a good amount that covers as much of the overlap as possible. Then you need to click on one model, then click on the same place on the other model. ![]() As I say, it doesn’t matter, so I just select the top one and hit ‘ok’. That’s good, because as you can see I didn’t properly rename the meshes in the Tree, only their parents, so I can’t actually tell which is which. Now, it doesn’t really matter here because neither is in a ‘true’ position. That brings up a window asking which model we want to be the reference: So with both meshes cropped and side-by-side, we shift-select both meshes in the DB tree, then click the ‘roughly align two meshes’ button: ![]() The reason is we’ll be picking analogous points on both meshes, and if they are overlapping it’s hard to see and click on the correct place. ![]() To do this, I hide one mesh, then use the scissors tool to cut out just the part I want to keep: I’m very generous with the cutting – you really want to get rid of any large triangles and poorly meshed parts, as well as non-object mesh.īefore I try to align the meshes, I use the transform and rotate tool to just move them so they are next to each other and I can see them clearly: I’ve moved ‘down’ so that it’s next to ‘up’ and oriented the same way. Then we need to cut out the stuff we don’t want in the model – the white turntable, and any floating parts of the mesh. Here, I’m going to show you how to merge the two seperate scans using CloudCompare.įirst thing to do is load both models into CloudCompare: I’ve renamed the meshes once imported into CloudCompare, as you can see on the left. I’ve called the two models ‘up’ and ‘down’, for conveinience. Both models were made from 100 photos (50 DSLR, 50 Note 8) from my automated setup. When they work, they work really well, but sometimes they can be unreliable.įor this demonstration, I’ve got two photogrammetry scans of a fossil Trilobite, one of it lying on it’s lower surface, and one of it resting upside down. Both Agisoft Metashape and AliceVision Meshroom have functions for aligning two sets of photos, but they rely on quite a lot of overlap. In those cases, you can scan both sides separately then merge the meshes. Or something that’s an awkward shape for reliably balancing one way or another. Maybe you’ve got a large bone or other object that can’t be moved easily. This is the method my automated photogrammetry set-up relies on – the white turntable and white background mean I take photos with the object in multiple orientations, then throw all the photos in Meshroom in one go.īut that’s not always possible. With photogrammetry, that might involve what I’ve seen called the ‘void’ method – place your object against a featureless background, so that when you turn it over, the software doesn’t realise and things the camera has moved instead. Ideally, when you scan an object with laser scanning or photogrammetry, you want to get the whole thing in one go. This post will show you how to align a scan in two or more parts using CloudCompare.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |