Privacy Preserving Visual SLAM

  • 2020-07-20 18:00:06
  • Mikiya Shibuya, Shinya Sumikura, Ken Sakurada
  • 21

Abstract

This study proposes a privacy-preserving Visual SLAM framework for estimatingcamera poses and performing bundle adjustment with mixed line and point cloudsin real time. Previous studies have proposed localization methods to estimate acamera pose using a line-cloud map for a single image or a reconstructed pointcloud. These methods offer a scene privacy protection against the inversionattacks by converting a point cloud to a line cloud, which reconstruct thescene images from the point cloud. However, they are not directly applicable toa video sequence because they do not address computational efficiency. This isa critical issue to solve for estimating camera poses and performing bundleadjustment with mixed line and point clouds in real time. Moreover, there hasbeen no study on a method to optimize a line-cloud map of a server with a pointcloud reconstructed from a client video because any observation points on theimage coordinates are not available to prevent the inversion attacks, namelythe reversibility of the 3D lines. The experimental results with synthetic andreal data show that our Visual SLAM framework achieves the intendedprivacy-preserving formation and real-time performance using a line-cloud map.

 

Quick Read (beta)

loading the full paper ...