Loop closure detection
Source:
cometlabs
See also:
loop-closing-in-viorb
Loop closure
- Process of observing the same scene by non-adjacent frames and adding a constraint (relationship? association?) between them
- A long-term data association in the VSLAM Framework (part of front end)
- Sort of incorporates topological SLAM into metric SLAM
Importance
- Final refinement step (in data association)
- Important for obtaining a globally consistent SLAM solution, especially when optimising over a long period of time
Basic loop closure detection
Match the current frame to all previous frames using feature matching
- Computationally expensive, not always suitable for real-time applications
- For RT—a possible solution: Define key frames and perform the comparison with only the key frames
Problems with loop closure
- False positive/perceptual aliasing: Places are different, but are perceived as the same
- False negative: Places are the same, but perceived as being different
- Abhilfe: precision recall curve