An Economical Approach to 3D Scannning

Описание: All figures, tables, and graphs were created by the researchers unless otherwise cited

1. What is Neural Radiance Fields (NeRF)?
NeRF, developed by Mildenhall et al. in 2020, is a technology that turns 2D images into 3D Representation with Neural Networks.



2. What have our team done?
The NeRF algorithm's need for precise 5D data—position and orientation for each video frame—depends on structure-from-motion (SfM) and COLMAP. Yet, COLMAP struggles with tracking featureless objects, complicating scans of symmetrical or reflective items.

To address these issues, our team proposes a new 3D scanning framework featuring a robotic arm to stabilize object rotation and use forward kinematics for accurate camera positioning, eliminating dependence on COLMAP or SfM.



3. What have our team achieved?
3.1 Our prototype, costing under $100, produces competitive 3D scans despite slight imperfections, providing a cost- effective solution.

3.2 COLMAP SfM struggles with camera positioning for transparent objects, leading to poor mesh quality. In contrast, forward kinematics achieves precise, quick positioning regardless of the mesh's texture.