A New Era of Indoor Scene Reconstruction: A Survey
Hao Wang, Minghui Li
Abstract
Indoor scene reconstruction plays a pivotal role in the fields of computer graphics and vision, significantly impacting areas such as robotics and augmented reality. Recently, advancements in Neural Radiance Fields (NeRF) and 3D Gaussian Splashing (3DGS) technologies have notably enhanced scene rendering efficiency, accelerated the generation of images from new perspectives, and simplified tasks such as dynamic reconstruction, geometric editing, and physical simulation. Furthermore, the widespread application of consumer-grade RGB-D cameras has promoted the progress of indoor scene reconstruction technology. However, when dealing with objects that are reflective, highly lit, or transparent, as well as distant objects, the depth images produced by these cameras often contain considerable noise and incomplete data, posing challenges to existing systems. This survey comprehensively reviews the latest progress in indoor scene reconstruction using current technologies and consumer-grade cameras, covering aspects off depth image processing, camera pose estimation, surface fusion, optimization, and data completion. By comparing and analyzing the latest methods, this survey aims to provide beginners with an introduction to the field and offer experienced researchers a comprehensive technical overview, thereby fostering the continuous development of this field.