Google’s VPS – how well will it work?
May 19, 2017
Artificial Intelligence and Artificial Perception
December 4, 2017
Show all

Benefits of leveraging KudanSLAM with ARKit on iOS devices

Though ARKit's optimised IMU allows it to perform robustly, there are multiple features and performances which can be improved upon. Through leveraging the visual-based tracking of KudanSLAM with the IMU sensing of ARKit, the total performance can be enhanced beyond its usual capabilities.

  1. Feature and Performance analysis of ARKit:

    • Uses visual inertial odometry, not SLAM
    • Provides accurate 6DOF tracking even with the camera covered, thanks to good use of a well calibrated IMU
    • Tracking is driven mainly by the IMU, with vision used primarily for correcting drift
    • Creates and tracks a set of sparse feature points with a single camera (monocular) setup
    • New feature points are created via translational movement. Pure rotation does not create new feature points
    • Points seem to have a short lifetime of a few seconds but can be recreated again upon movement
    • Points are prone to drift, and often float or have inaccurate triangulation
    • Low saliency of points
    • High framerate camera (min. 60fps) makes point tracking easier due to smaller distances moved frame to frame
    • No noticeable rolling shutter effect, likely due to the high framerate camera combined with data from the IMU to perform corrections
    • Fairly low CPU usage (40-50%) on iPhone 7 and later
  2. KudanSLAM on top of ARKit:

    • KudanSLAM can map using odometry data from ARKit
    • Maps can be saved and loaded, allowing relocalisation of pre-mapped areas and map sharing, etc.
    • Maps can be used to correct the drift present in ARKit
    • Provides loop closure for higher accuracy tracking and mapping
    • Produces a denser and more accurate point cloud of the area
  3. KudanSLAM components as part of ARKit (in addition to the above):

    • Improved triangulation of points
    • Accurate tracking of points
    • Lower power consumption

Comments are closed.