Automatic observation for 3D reconstruction of unknown objects using visual servoing

This paper presents a method for the automatic observation of unknown objects. We aim at finding the position of the targeted object and at capturing multiple views of its shape using a single eye-in-hand camera. The main goal is modeling the unknown object via 3D reconstruction before grasping and manipulating it with a robot hand. The proposed method built over a visual servo loop uses simple features to constantly gaze at the object and handles a three-step exploration with the same control structure. Images captured with the experimental setup are processed online to recover a model with millimetric accuracy.