keyframe selection
**Keyframe selection** is the **policy that chooses which frames become map anchors to balance information richness, computational cost, and map size** - selecting informative keyframes is essential for stable SLAM and efficient optimization.
**What Is Keyframe Selection?**
- **Definition**: Decide when to insert a new keyframe based on motion, overlap, and tracking quality criteria.
- **Purpose**: Avoid storing redundant frames while preserving enough coverage for relocalization and mapping.
- **Inputs**: Pose change, feature novelty, uncertainty, and scene dynamics.
- **Outputs**: Sparse set of representative frames used in map and backend optimization.
**Why Keyframe Selection Matters**
- **Efficiency**: Fewer redundant keyframes reduce memory and compute burden.
- **Optimization Quality**: Better keyframe distribution improves graph conditioning.
- **Relocalization Strength**: Representative landmarks increase successful place matching.
- **Real-Time Performance**: Controls backend workload growth over long missions.
- **Map Longevity**: Good keyframe policies support robust long-term operation.
**Selection Strategies**
**Motion Thresholding**:
- Insert keyframe after sufficient translation or rotation.
- Simple and effective baseline.
**Information Gain**:
- Add keyframe when new observations provide significant scene novelty.
- Reduces overlap redundancy.
**Quality-Aware Policy**:
- Trigger keyframe when tracking uncertainty rises.
- Improves robustness in difficult segments.
**How It Works**
**Step 1**:
- Evaluate current frame against latest keyframe using motion and overlap metrics.
**Step 2**:
- Insert frame as keyframe if thresholds or uncertainty rules are satisfied; otherwise continue tracking.
Keyframe selection is **the data-budget control mechanism that keeps SLAM maps informative without becoming computationally unmanageable** - careful policy design improves both speed and global accuracy.