3D Structure Estimation of Room Environment Using Semantic Segmentation and Model Fitting

Recent advances in VR devices and technology have increased the demand for technologies that reflect real space in virtual space. In particular, technologies that reflect the indoor environment in a virtual space are used in various fields such as architecture, interior design, and robotics. To reflect a real space in a virtual space, estimating the 3D structure of the real space is significant. Yet, this requires expensive equipment and poses challenges in computational complexity and ease of measurement. In this study, we propose a method of estimating 3D structure of indoor environments using a single image. Existing methods generally use depth estimation, but in this study, we introduce a new approach using semantic segmentation and rectangular fitting. Experimental results with 3D CG show that the proposed method outperforms existing methods in accuracy and competitive results.