ExRealis

Geometry registration. Devices requiring multiple 3D acquisitions from different viewpoints (structured light range scanners, lidars, etc) provide for each acquisition an independent point cloud, which needs to be realigned with the others, similarly to a big 3D puzzle. An algorithm derived from my PhD. works enabled us to automate this process for the specific case of structured light range scanners, but ExRealis also provides manual registration thanks to a dedicated user interface, as well as registration refinement by iterative minimization algorithms.

Cleaning. Since no scanning technology is now noise free, it is not uncommon to recover erroneous measured points, which risk to disturb the following processing. Here again, a dedicated user interface enables manual cleaning. However, this is often a tedious task. That’s why heuristics have been implemented in order to automate this process, using confidence values computed from geometric or colorimetric criteria.

Surface reconstruction. Point clouds provided by shape measurement technologies constitute a representation which is not adapted to many usages to which 3D models are intended (realistic rendering, physical simulation, etc). It is thus often necessary to obtain a closed description of the surface, in the form of a 3D mesh. To do so, we integrated to ExRealis some methods derived from scientific literature to extract triangulated surfaces from sets of measured points.

Surface simplification. To display 3D models on devices with low graphical performances (smartphones, tablets, autonomous virtual reality head mounted devices), it is important to be able to reduce geometry complexity. ExRealis incorporates different algorithms allowing this reduction while limiting the loss of geometry details, so as to preserve as most as possible the object shape during the simplification step.

Surface parametrisation. Giving an appearance to a 3D object is classically made through the use of textures, which are images that may contain various informations (albedo, reflectivity, roughness, details, and so on). To make each portion of the surface corresponding to an image part, the 3D object first need to be unfolded and flattened onto a plane, similarly to the earth globe onto a planisphere. Several algorithms to solve this so-called parametrisation step are available in the software.

Photography registration. To reconstruct the colour of a model from pictures, a first step consists in recovering the viewpoints from which they have been shot, so as to be able to project their content onto the 3D model, similarly to what a video-projector would do. Camera calibration methods have then been integrated to ExRealis for this purpose, as a preliminary step to appearance reconstruction.

Colour map reconstruction. The use of multiple pictures for colour map reconstruction requires to manage adequately the surface areas where several images are projected. Indeed, in these overlapping regions, a naive blending may result in visual artefacts. ExRealis provides more elaborated blending strategies based on visibility criteria which allow, for instance, to reduce the impact or reprojection errors or to eliminate the focus blur that may appear in some of the source images.

Normal map reconstruction. Beyond a certain threshold, surface simplification may lead to the loss of geometry details. By storing those details into a normal map, it thus becomes possible to modify illumination during the rendering, in order to simulate geometry complexity over simplified 3D meshes. This approach is much cheaper than directly displaying highly detailed meshes. ExRealis allows the reconstruction of normal maps for coarse meshes from more detailed versions of those.