A Hybrid Learning-Driven Computer Vision Framework for Reverse Engineering via Enhanced 3D Shape Reconstruction

Reverse engineering (RE) has played a key role in producing low demands parts, especially with the recent advent of robust additive manufacturing (AM) techniques. The synergetic interaction of both cutting-edge RE and AM techniques significantly enhance part re-producing and minimize the product development cycle time, even if there is no blueprint for the desired product. Recently, computer vision algorithms have enhanced the RE process and strengthen its capabilities to reconstruct challenging shapes. Nevertheless, the large body of the reported literature is restricted to estimate the 3D shape of the scanned part from a single/multiple 2D/3D image based on predefined classes using supervised learning. The ability to reconstruct intricate geometrical features of real mechanical parts and complex shapes has not been fully realized yet. In this context, this paper reports on a hybrid learning technique-based conceptual computer vision framework to enhance RE process for reproducing of low demand products. The hybrid learning proposed herein is a supervised and unsupervised learning technique using a dual deep learning models to enrich the computer vision technique with the ability to reconstruct 3D complex features using a single 3D depth image.