3D Reconstruction of Unknown Objects

Performing tasks in unstructured environments is a challenging robotic research field. In applications like object fine manipulation, it is necessary that the object’s surface is known to plan the coordinated movements of the robotic fingers in a proper way.

In the literature, some techniques have been adopted to reconstruct the shape of the unknown object to be manipulated: they span from superficial to volumetric object reconstruction, employing visual techniques which are often complex from a computational point of view. In those applications where the object reconstruction has to be performed in real-time, the method proposed here and here is without doubt a good and competitive choice. In the proposed algorithm, a robot equipped with a calibrated camera, mounted in an eye-in-hand configuration, follows some trajectories around the object to be reconstructed. Several images are acquired along these paths. After some standard image processing operations, an ellipsoid is virtually placed around the object, and it is sampled by points, in turn, interconnected through virtual springs and dampers. These points dynamically shrink toward the ellipsoid’s center and stop when they intercept the so-called visual-hull. In this way, it is possible to reconstruct in a fast and accurate way the surfaces of the objects in unstructured environments.