sparse mapping
**Sparse mapping** is the **SLAM and SfM representation that stores selected salient landmarks instead of full surfaces to prioritize localization efficiency** - it focuses on distinctive points and descriptors that are reliable for pose estimation.
**What Is Sparse Mapping?**
- **Definition**: Build map from sparse set of 3D feature points and associated observations.
- **Landmark Type**: Corners, edges, and textured keypoints with robust descriptors.
- **Primary Goal**: Support accurate tracking and relocalization with low compute.
- **Typical Outputs**: Sparse point cloud, keyframe graph, and descriptor database.
**Why Sparse Mapping Matters**
- **Computational Efficiency**: Lower memory and optimization costs than dense maps.
- **Real-Time Readiness**: Suitable for embedded and resource-constrained platforms.
- **Robust Localization**: Distinctive landmarks provide stable pose constraints.
- **Scalable Operation**: Easier long-term map maintenance across large trajectories.
- **Backend Compatibility**: Works well with bundle adjustment and pose graph optimization.
**Sparse Mapping Pipeline**
**Feature Extraction**:
- Detect repeatable keypoints and compute descriptors per frame.
- Filter unstable points and outliers.
**Triangulation and Map Update**:
- Triangulate landmarks from matched observations.
- Insert into map with uncertainty tracking.
**Map Management**:
- Prune weak landmarks and redundant keyframes.
- Keep map compact and informative.
**How It Works**
**Step 1**:
- Match features across frames and estimate camera poses.
**Step 2**:
- Triangulate sparse landmarks, optimize map, and use descriptors for relocalization.
Sparse mapping is **the efficiency-oriented map representation that powers reliable localization with minimal geometric overhead** - it remains the default backbone in many real-time SLAM deployments.