Deep InsightGeometry · Perception · Robotics
← Knowledge map

03 / Perception

3D Perception

From pixels and point clouds to spatial representations a machine can reason about.

Series

SeriesImage-Based 3D ReconstructionRecovering camera poses and scene geometry and appearance from a set of photographs. The series follows the classical pipeline chapter by chapter, from the imaging model and epipolar geometry through feature matching, stereo, structure from motion, dense reconstruction, surface reconstruction and texturing.8 chapters

Published notes

Notes for this stop are being written — coming soon.

Questions on the roadmap

  1. Pinhole cameras and depthHow does a pixel get back into 3D space?
  2. The skeleton of SLAMWhat front end, back end and loop closure each solve.
  3. How should a world be stored?Trade-offs between point clouds, voxels, TSDF, occupancy grids and Gaussian splats.