|M.Sc Student||Holdstein Yaron|
|Subject||Reconstruction of Freeform Surface Objects Using Neural|
|Department||Department of Mechanical Engineering||Supervisor||Professor Anath Fischer|
Reverse Engineering (RE) is the process of reconstructing a computerized model from a digitized object. In recent years, RE has begun to play a major role in CAD systems for acquiring digitized models of real 3D complex objects. However, common reconstruction methods suffer from accuracy problems and cannot guarantee topology preservation.
The Neural Network method is a relatively new method in RE that has the potential of achieving the desired outcome. A Neural Network (NN), which resembles an artificial intelligence (AI) algorithm, is a set of interconnected neurons, where each neuron is capable of making an autonomic arithmetic calculation. Moreover, each neuron is affected by its surrounding neurons through the structure of the network.
This research first explores a Growing Cell Structure Neural Network (GCS NN) with tetrahedral structure for volumetric reconstruction. Then, it proposes a new approach that utilizes the Growing Neural Gas Neural Network (GNG NN) technique to reconstruct a triangular manifold mesh. The second method has the advantage of reconstructing the surface of a freeform object of n-genus without a priori knowledge regarding the original object. The resulting mesh can be used by a subdivision scheme for smoother results. Another application examined in this research was multiresolution representation using GNG-based technique. The proposed Neural Network based method produces a high quality manifold triangular mesh. The feasibility of the proposed method and its applications are demonstrated on several examples of freeform objects with arbitrary topologies.