Indoor 3D geometry estimation using single panorama image

A 3D layout provides a comprehensive representation of spatial arrangements within a room, offering insights critical for architectural planning, augmented reality, and various computer vision applications. Panorama images, capturing a 360-degree view, hold immense potential for scene understanding due to their immersive nature. However, extracting detailed 3D layouts from such images remains a complex task. To overcome this, we propose a novel algorithm for the estimation of 3D indoor layouts using a single panorama image. Our contributions are dual-fold. Firstly, we introduce the ResNet-50 based corner detection algorithm, which empowers precise corner localization within the confines of a single panorama image. This algorithm’s versatility is evident in its effectiveness across a wide spectrum of indoor layouts, accommodating diverse architectural designs such as cuboid and non-cuboid rooms. Secondly, through comprehensive experimental validation, we highlight the algorithm’s superior performance in terms of accuracy, underlining its potential in refining scene comprehension within panorama images. Furthermore, by integrating the corner detection algorithm with a 3D ray tracing technique, we establish an integrated solution for reconstructing 3D indoor layouts. This integrated approach offers a holistic solution for translating the detected corners into an accurate 3D layout reconstruction. Our results validate the algorithm’s impressive accuracy and speed, advocating for its applicability in practical scenarios. With its good accuracy and speed, this approach holds great promise for advancing both scene understanding and 3D modeling within the exciting realm of panorama-centric indoor environments.