relocalization
**Relocalization** is the **SLAM recovery process that estimates current pose after tracking failure by matching the live view against a previously built map** - it allows robots to resume operation after occlusion, rapid motion, or temporary sensor degradation.
**What Is Relocalization?**
- **Definition**: Re-estimating absolute camera or robot pose in a known map when local tracker is lost.
- **Trigger Events**: Motion blur, feature starvation, abrupt viewpoint change, or temporary sensor outage.
- **Input Signals**: Current frame descriptors, map keyframes, and geometric verification constraints.
- **Output**: Recovered pose with confidence, then re-entry into normal tracking loop.
**Why Relocalization Matters**
- **Operational Continuity**: Prevents full system restart when tracking breaks.
- **Safety**: Critical for robots and autonomous systems in dynamic environments.
- **Map Reuse**: Leverages prior mapping investment across repeated runs.
- **Drift Mitigation**: Anchors pose back to globally consistent map coordinates.
- **User Experience**: Improves robustness in AR and navigation products.
**Relocalization Pipeline**
**Place Retrieval**:
- Query current observation against keyframe database.
- Return candidate map locations by descriptor similarity.
**Geometric Verification**:
- Match feature correspondences and solve PnP or scan alignment.
- Reject false positives from perceptual aliasing.
**Tracking Reinitialization**:
- Resume local tracking from recovered pose.
- Update uncertainty and map consistency state.
**How It Works**
**Step 1**:
- Detect tracking loss and run fast global place search over stored map descriptors.
**Step 2**:
- Verify best candidate geometrically, estimate pose, and hand control back to tracker.
Relocalization is **the recovery mechanism that turns SLAM from fragile short-term tracking into persistent long-term autonomy** - robust place retrieval plus geometric verification is the key to reliable restart behavior.