Initialisation of monocular SLAM

Source:  Lamarca 2019 DefSLAM

Depth information has to be generated before localisation can be performed — how?

• Capture multiple images which have enough parallax
• These images with parallax allows depth information to be calculated (this uses motion parallax )
  • From these images, the map can be generated
  • Localisation can then be carried out with respect to the map (as long as camera doesn’t move off to an unexplored region)