Realtime Simultaneous Localization and Mapping - on Pixel phone

less than 1 minute read

Published:

Simultaneous Localization and Mapping (SLAM) answers two questions at once: where is the camera? and what does the world around it look like? This demo runs both in realtime.

How it works

  1. Feature tracking: detect keypoints in each incoming frame and match them against the previous frames.
  2. Pose estimation: estimate the camera motion from the matched features.
  3. Mapping: triangulate the matched features into 3D landmarks and add them to the map.
  4. Optimization: refine the camera poses and landmarks together to reduce accumulated drift.

The result is a camera trajectory and a sparse 3D map, built frame by frame as the video plays.