A Part Assembly Framework for Recovering 3D Geometry and Structure of Everyday Objects

Description

3D scanning is one of the most effective ways for 3D modeling, which is at the heart of computer graphics and visualization. The growing popularity of fast, easy-to-use, low-cost depth cameras (e.g., the Microsoft Kinect) presents new acquisition possibilities and makes it possible for everyone to digitize 3D objects at home. However, the resulting raw scans are often highly noisy and incomplete. Hence, a faithful reconstruction of the underlying 3D geometry is rather challenging even for the state-of-the-art techniques, let alone understanding its semantic structure.This work gets inspirations from the recent advancement of assembly-based 3D modeling, which is designed for artist-driven, open-ended 3D modeling. Our key idea is to let a computer (serving as a virtual artist) utilize acquired scans from depth cameras to automatically guide and constrain a process of part assembly. This poses an interesting but challenging research problem given the exponentially growing space by shape synthesis via part composition. We present a promising framework by heavily exploiting geometric priors for individual parts and their mutual relations exhibited in the shape repository. It results in a novel approach for recovering high-detailed 3D geometry from low-quality scan acquisition of everyday objects. Unlike existing surface reconstruction technologies, which are limited to low-level geometry reconstruction, our framework naturally enables high-level shape understanding and allows the creation of new semantic structures with respect to the acquired data. The produced structures can directly benefit applications like structure-aware shape editing and synthesis. In short, the proposed framework will turn consumer-level depth cameras to powerful personal 3D scanners and allow quick conversion of low-quality object scans to high-quality 3D digital models with semantic structures. The output of our framework can be directly fed into a wide range of applications, for example, semantic indoor modeling, augmented reality, reverse engineering and rapid prototyping etc.We are confident that we can complete the project and deliver the expected outcomes. We have had extensive research experiences in the field of shape analysis, which we believe have paved the way for the proposed ideas. We expect that the new project will follow the success of our current projects and make continuous contributions to both research and industry.