Simultaneous reconstruction of 3D geometry and texture map using a single RGB-D camera

What is it about?

3D reconstruction is a long-standing important research topic to digitalize real 3D objects or scenes. This technique could be used for various applications, such as augmented reality (AR) and virtual reality (VR). We propose a novel framework that simultaneously reconstructs 3D geometry and texture map of target scenes or objects in real time.

Featured Image

Photo by Hemant Kanojiya on Unsplash

Photo by Hemant Kanojiya on Unsplash

Why is it important?

Prior real-time 3D reconstruction methods using a single RGB-D camera have used voxel color representation. However, this representation essentially requires a high-resolution voxel grid to represent a high-quality color of reconstructions, resulting in a high memory footprint and computational amount. In addition, in spite of such costs, existing work produces blurry color due to various error sources, including inaccurate camera pose estimation. To resolve these problems, we, for the first time, introduce the texture map representation into a real-time 3D reconstruction pipeline, and also provide photometric optimization to resolve color misalignments. By doing so, we can obtain sharp textured 3D models immediately after scanning without any post-processing.

Perspectives