Multi-Scale Voxel Hashing and Efficient 3D Representation for Mobile Augmented Reality

Yi Xu, Yuzhang Wu, Hui Zhou; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 1505-1512

Abstract


In recent years, Visual-Inertial Odometry (VIO) technologies have been making great strides in both research community and industry. With the development of ARKit and ARCore, mobile Augmented Reality (AR) applications have become popular. However, collision detection and avoidance is largely un-addressed with these applications. In this paper, we present an efficient multi-scale voxel hashing algorithm for representing a 3D environment using a set of multi-scale voxels. The input to our algorithm is the 3D point cloud generated by a VIO system (e.g., ARKit). We show that our method can process the 3D points and convert them into multi-scale 3D representation in real time, while maintaining a small memory footprint. The 3D representation can be used to efficiently detect collision between digital objects and real objects in an environment in AR applications.

Related Material


[pdf] [supp]
[bibtex]
@InProceedings{Xu_2018_CVPR_Workshops,
author = {Xu, Yi and Wu, Yuzhang and Zhou, Hui},
title = {Multi-Scale Voxel Hashing and Efficient 3D Representation for Mobile Augmented Reality},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}