Fitting Deformable Mesh Models to Raw Point Clouds

In this chapter, we are going to discuss a project for using deformable mesh models for fitting raw point cloud observations potentially coming from a raw depth camera sensing result. Raw point cloud observations from depth cameras are usually in the format of point clouds without any information on how these points are connected; that is, the point clouds don’t contain information about how surfaces can be formed from the points. This is contrary to a mesh, where the list of faces defined by the mesh shows us how the surfaces are. Such information on how points can be gathered into surfaces is important for downstream postprocessing, such as denoising and object detection. For example, if one point is isolated without any connection to any of the other points, then the point may likely be a false detection by the sensor.

Thus, reconstructing the surface information from point clouds is usually a standard step in 3D data...