Hi, in this lesson, we will be talking about the 3D reconstruction of real objects, starting from photos in practice. The previous lesson was about the theory and methods of 3D survey in cultural heritage. In the current, in the current lesson, sorry, the 3D survey in practice using an image-based approach, how does it work? We will take some photos on, on the field. We use some software tools. We obtain a point cloud, and we use it to make our final 3D model. Okay, let's see how in practice here. [NOISE] Okay, I take my old iPhone 4, and, let's start, our 3D survey of, the statue [SOUND] Okay. A set of photos useful for our 3D reconstruction. This is a, a very interesting statue. It is from the Cucuteni culture, nowadays in Romania in Europe, estern Europe and is a, a reproduction, a clay reproduction of a statue from the fourth millennium, before Christ. Okay, now let's start to make our 3D model. The first software we will use is the VisualSFM. That is, stands for structure for motion. The Visual Structure for Motion is a free for personal, non-profit or academic use, and it's perfect for our purpose of this course. You can download it for every OS, but I warmly recommend Windows version because it's easier to be installed. But it's fine also for Linux, Mac OS X, and so on so forth. First of all, we open our VisualSFM folder, and start the VisualSFM.exe file. Here we get this kind of visualization, the VisualSFM is a software with a blank window, and we can import our data set, our files, the photos just took with the iPhone 4. So, we have to follow, to open this way open more from the file selection. Menu and, or through this icon. In the GUI go to the folder and we select all the files present. Now the, the VisualSFM software is charging the images inside the canvas. Okay, well the photos are now inside VisualSFM. First of all, we have to compute the missing matches between these photos with this icons. With this tool, VisualSFM finds for homologous points between two or more photos I mean. Between this and this photo, there are similarities. The same points on the these two photos, for examples, the eyes of the statue that are the same in one photo and the second photo. With this command, we get some extra files in our dataset and for each JPEG photos, we obtain a .sift file, I mean, with the, in this sift file, there, are written all the other photos similar to the current one. So, this is the photo and in this file we have all the other photos that, that there are similar to this one. Okay. The miss, miss, the missing matching point finished, in some minutes. Okay, and now we have the computed 3D reconstruction with this icon here, and what is going on? VisualSFM is reading all the information about the. the similarities between the photos and is drawing on the 3D scene. The relative position of these photos in the space and it computes the so-called sparse points. Here we have the representation in 3D of the homologous points between the photos of the dataset. So it is not a very accurate point cloud. It is just a starting point for.. ..to have later the dense point cloud, the accurate one. [NOISE] Okay, now summing up of this first part. The steps involved, the Sift process, we use all the photos in the, we take all the photos in the VisualSFM and we look for homologous points between photos. So, there are an unordered series of photos, and we, and the software find out the homologous points between these photos. This photo and this photo are one next the other one, and so on, and the bundle the extraction of, of the position, the relative position of the cameras. In the scene and in the end the creation of the dense point cloud. Okay. In this course we use VisualSFM but there are other software we can use. There are free open source software or pay software, there is Apero-MicMac that is a that has a photogrammetric approach very accurate. And you can use it freely also for commercial projects and there is also the Python Photogrammetry Toolbox. The same license, or there is a, a paid software like the Agisoft photo scanner. It works very well, or instead of scan and so on and so forth. Okay. Now, the second software we will use Meshlab. It's open source, and it's, so it's free to use also for commercial. Is developed by CNR-ISTI Pisa, Italy and is very common software, is used worldwide. With this software you can edit and make 3D content from real objects, and, with great level of accuracy with the state of the art algorithms in 3D computer graphic. You can download it for every OS, Linux, Windows, and MacOS and so on. Now we have the dense point cloud and we need to open it with the MeshLab software. We search it on, on Google and the first one result is MeshLab, and here we can load it for Windows, for Mac OS X and for Linux you have.. ..in the repository of your distro. Here we download the exe, the Mesh Lab is free, so we have no problem and in seconds we get a free copy. Now we have opened, now we opened the Mesh Lab too and we, we open project file the VisualSFM reconstruction is a project for Mesh Lab where we can import the sparse point cloud. So the not very accurate point cloud. with all the homologous points we turn off the light from here and we can zoom in with the middle mouse button scroll to see. the statue Okay, now we open the layer button here. To see to have a look a the content of this project we have one point cloud, and here several photos connected to this point cloud. They are the photos I just took, with my iPhone. Now we have to import, the dense point cloud. We can import the dense point cloud from the file menu, import mesh, and we can find. Our point cloud just inside the Cucuteni folder cucuteni statue .ply this is the format used. Okay. So now we have two different. Point clouds. One dense and one sparse. We need to work on the dense one. So we have to uncheck the eye of the spare one and work just on the dense one. First of all we have a lot of stuff that we don't need. So all the street, outside where I took the photos for you. For the exercise. It's better outside for the light, and for the characterization of the photos. We have a lot of feature in the photos. There are stones of the streets the walls of the street and so on, and the photogrammetry works better with a lot of objects inside the scene, and we have to clean this point cloud. through trashing all the point we don't need at all. So, we have some tools inside Mesh lab. Here is the selection, window selection, and here we have the, point delete. These two commands. [SOUND] We have no we have only a dense point cloud of our statue but we need a surface of our statue. Something to to print or to paint over the color. So we select the filter menu and remeshing simplification and reconstruction. We, we select the surface reconstruction with the algorithm named Poisson so with this tool, we can obtain the surface of the statue starting from this dense point cloud. The parameters you can use are twelve and ten for the octree depth how much resolution we have to set in this 3D statue Okay. We apply this and calculate. And here's what you get. A mesh, a poisson mesh, obtained from the dense point cloud. This is a blank mesh so we need some color. And we have several photos of this statue and we can start from them to obtain the color of the, the model and from the filter menu. The texture menu we can obtain the parametrization plus texturing from registered raster The registered rasters are the photos. So we set a big texture of 4,096. a texture name will be texture.png And we check this, all these fields, color correction, and so on. And we apply to obtain our color from photos. Okay. And now we obtain our 3D model of the statue. We can, improve this model with other steps of refinement, in other 3D software like Blender 3D. Or just inside the Meshlab. But for the sake of simplicity. You will find out that these are just on the complete video you can access on the course website on in our course. Have a look to the export of this this statue. We can export it in the OBJ format from the file menu export mesh OBJ format and choose setting options, we have to, set the text coordinate and, to be sure that there is the texture dot png, file. And that's all. Summing up this part of lesson. The steps involved in mesh lab are cleaning the dense point cloud. Creating the mesh with the poisson filter. And creating the textures from the photos. And the export of this model to the 3rd and last software.