Spatial Analysis of 3D Space using 2D Images

For a human being, the process of perceiving any object or scene in three dimensions, i.e., its length, breadth and height, seems to be a very simple and obvious task. But this same task of perceiving any object in three dimensions is a very complicated and taxing process for a machine. The objective of the project is the creation of a linked 3D model of two rooms, with the training and testing inputs to the neural network being real-life 2D images of various rooms, taken from different points of view or panoramic images. This aim is an extension of the previously published technical papers (GQN, U-Net, ResNet, RoomNet, LayoutNet and HorizonNet) that deal with 3D model creation of only one room (without any room linking) using only virtual images available on the Internet (no real-life images used for training or testing the neural network). The programming language used is Python, utilizing the Jupyter Notebook (a programming language editor and compiler) in Anaconda Integrated Development Environment (Anaconda I.D.E.).

Keywords: perceiving, training, testing, real-life, panoramic, virtual

Suggested Citation: Suggested Citation

Benjamin, Annie and Kanase, Pradhumn and Likhite, Maitreyee and Noronha, Noella and Jadhav, Anita, Spatial Analysis of 3D Space using 2D Images (April 8, 2020). Proceedings of the 3rd International Conference on Advances in Science & Technology (ICAST) 2020, Available at SSRN: https://ssrn.com/abstract=3586882 or http://dx.doi.org/10.2139/ssrn.3586882