Three-dimensional (3D) object reconstruction is\nthe process of building a 3D model of a real object. This\ntask is performed by taking several scans of an object from\ndifferent locations (views). Due to the limited field of\nview of the sensor and the object�s self-occlusions, it is a\ndifficult problem to solve. In addition, sensor positioning\nby robots is not perfect, making the actual view different\nfrom the expected one. We propose a next best view\n(NBV) algorithm that determines each view to reconstruct\nan arbitrary object. Furthermore, we propose a method\nto deal with the uncertainty in sensor positioning. The\nalgorithm fulfills all the constraints of a reconstruction\nprocess, such as new information, positioning constraints,\nsensing constraints and registration constraints. Moreover,\nit improves the scan�s quality and reduces the navigation\ndistance. The algorithm is based on a search-based\nparadigm where a set of candidate views is generated\nand then each candidate view is evaluated to determine\nwhich one is the best. To deal with positioning uncertainty,\nwe propose a second stage which re-evaluates the views\naccording to their neighbours, such that the best view is\nthat which is within a region of the good views. The results\nof simulation and comparisons with previous approaches\nare presented.
Loading....