Towards Scalable Multi-View Reconstruction of Geometry and Materials

Area of Science:

• Computer Vision
• Computer Graphics
• 3D Reconstruction

Background:

• Estimating 3D scene properties like geometry and materials from captured data is crucial for various applications.
• Existing methods struggle with large-scale scenes or require specialized, stationary capture setups.
• Mobile capture systems offer flexibility but present challenges in data processing and reconstruction accuracy.

Purpose of the Study:

• To develop a novel, scalable method for joint recovery of camera pose, object geometry, and spatially-varying Bidirectional Reflectance Distribution Function (svBRDF).
• To enable reconstruction of 3D scenes that exceed object-scale using mobile, hand-held capture systems.
• To provide a unified framework for accurate and globally consistent 3D scene modeling.

Main Methods:

• Utilizing high-resolution RGB-D images from a mobile, hand-held system with active point light illumination.
• Formulating the joint estimation problem within a single objective function minimized by gradient-based solvers.
• Implementing a distributed optimization algorithm for reconstructing 2.5D keyframe-based scene representations.
• Introducing a multi-view consistency regularizer for seamless integration of local optimizations into a global model.

Main Results:

• Demonstrated favorable comparison against existing baseline methods.
• Successfully reconstructed diverse objects and materials with high accuracy.
• Showcased the scalability of the method for spatially larger scenes and a large number of observation views.
• Validated the importance of individual components within the proposed formulation.

Conclusions:

• The proposed method offers a significant advancement in scalable geometry and material estimation from hand-held scanners.
• The approach facilitates the creation of globally consistent 3D models from complex, large-scale scenes.
• This work paves the way for more accessible and accurate 3D content creation using mobile capture technologies.