3D Environment Understanding in Real-time Using Input CAD Models for AR Applications

Abstract

The automatic recognition of real environments has become a relevant issue for multiple purposes in computer graphics, computer vision and artificial intelligence. In this work, we focus on environment understanding according to input CAD models for an Augmented Reality (AR) application. We provide a novel solution for the management of building infrastructures in indoor spaces. The use case has been the University of Jaén to visualize correctly their service facilites from AR. To this end, firstly, the CAD models (2D) have been segmented in order to simplify its geometry. As a result, an efficient data structure has been created for real-time alignement to scanned data. Secondly, we have developed a mobile application based on ARCore library to capture and generate 3D planes of the user environment. Finally, we have carried out a method to align automatically the virtual elements such as walls, doors and grounds to the real world. The main objective of this research is to calculate the needed geometric transformations of virtual elements and thus, to achieve a correct overlappping with the real world by understanding their physical and spatial constraints in real time.