Classification of image-based camera localization approaches
Parent: SLAM Index Source: [Wu 2018 Image-based camera localization: an overview](wu 2018-image-based-camera-localization_-an-overview.md)
Unknown environment (must be reconstructed from image data) Online/real-time mapping (SLAM)
- geometric metric SLAM
(accurate computations, therefore still widely used in practice)
- monocular
- multiocular
- multi-kind sensor
(active)
- loosely-coupled
- closely-coupled
- learning SLAM (active)
- needs a prior dataset to train NN – dataset determines performanace of the SLAM
- low generalisation capability, therefore not as flexible as geom. SLAM
- Has a 3D map
- topological SLAM (getting more unpopular)
- marker SLAM (semi-known environment)
[GEOM] Filter-based SLAM [GEOM] Keyframe-based SLAM(active)
- Feature-based
- Direct