3D Reconstruction of objects from RGB images and depth information using deep learning

Abstract

Object reconstruction is the manner of producing a computer model of the 3D appearance of an object from two-dimensional photos. It's the opposite procedure of obtaining 2D photos from 3D scenes. 3D reconstruction of objects from their digital pictures is a time-efficient and convenient manner of analysing the structural features of the item being modelled. Currently there may be an essential need for 3D content for computer graphics, virtual reality and communication, triggering an alternate emphasis for the requirements. Many present methods for constructing 3D objects are built round specialized hardware resulting in a high fee, information scanning barriers due to environment conditions which can't satisfy the requirement of its new programs. The art of three-dimensional reconstruction of objects and scenes has been a broadly researched topic. In this Master’s thesis, I proposed to address the above problems by developing a Deep Learning approach to reconstruct the object. This type of approach does not depend too much on the environment condition and the cost is low. However, the proposed method mostly targets the reconstruction of objects other than reconstruction of scenes. This research attempts to develop a Deep Learning based 3D reconstruction method for objects to avoid the limitations of the current 3D reconstruction approaches.