(Wu 2018) Image-based camera localization

Authors: Wu, Tang, Li

Abstract/Contents

• overview (classification) of image-based camera localization
• classification of image-based camera localization approaches
• techniques, trends
• only considers 2D cameras
• focuses on points as features in images (not lines etc)

Chapters

• Classification of image-based camera localization approaches
• Multisensor fusion  —  why use the visual-inertial sensor combination?
• Loose vs Tight coupling
• Filter localisation methods
• Some optimisation-based tightly-coupled multisensor SLAM algorithms

Questions

• What’s a metric map – normal map (with landmarks, normal distances) as opposed to a topological one

Takeaway

• Learning SLAM is gaining in popularity, but geometric SLAM is often the chosen method for most applications, as it is more generalisable and at the same time reasonably accurate
• For reliability and low cost practical applications, multisensor vision-centred fusion is an effective method.
• Possibly interesting: embedded SLAM algorithms
• Trend for a practical SLAM system: integrating all possible techniques
  • e.g. geometric and learning fusion, multi-sensor fusion, multi-feature fusion, feature based and direct approaches fusion
  • may solve the current challenges of poorly textured scenes, large illumination changes, repetitive textures, highly dynamic motions