Analysis and Modeling of 3D Indoor Scenes (1706.09577v1)

Abstract: We live in a 3D world, performing activities and interacting with objects in the indoor environments everyday. Indoor scenes are the most familiar and essential environments in everyone's life. In the virtual world, 3D indoor scenes are also ubiquitous in 3D games and interior design. With the fast development of VR/AR devices and the emerging applications, the demand of realistic 3D indoor scenes keeps growing rapidly. Currently, designing detailed 3D indoor scenes requires proficient 3D designing and modeling skills and is often time-consuming. For novice users, creating realistic and complex 3D indoor scenes is even more difficult and challenging. Many efforts have been made in different research communities, e.g. computer graphics, vision and robotics, to capture, analyze and generate the 3D indoor data. This report mainly focuses on the recent research progress in graphics on geometry, structure and semantic analysis of 3D indoor data and different modeling techniques for creating plausible and realistic indoor scenes. We first review works on understanding and semantic modeling of scenes from captured 3D data of the real world. Then, we focus on the virtual scenes composed of 3D CAD models and study methods for 3D scene analysis and processing. After that, we survey various modeling paradigms for creating 3D indoor scenes and investigate human-centric scene analysis and modeling, which bridge indoor scene studies of graphics, vision and robotics. At last, we discuss open problems in indoor scene processing that might bring interests to graphics and all related communities.