monocular slam
**Monocular SLAM** is the **visual SLAM variant that uses a single camera stream to estimate pose and reconstruct map structure** - it is lightweight and widely accessible, but must resolve scale ambiguity through motion and optimization.
**What Is Monocular SLAM?**
- **Definition**: SLAM using one RGB camera without direct depth measurements.
- **Primary Challenge**: Absolute scale is unobservable from single-view geometry alone.
- **Initialization Need**: Requires sufficient parallax to triangulate initial landmarks.
- **Common Systems**: ORB-SLAM family and direct monocular pipelines.
**Why Monocular SLAM Matters**
- **Hardware Simplicity**: Minimal sensor setup for low-cost deployment.
- **Wide Availability**: Works with commodity cameras on phones and robots.
- **Research Importance**: Strong baseline for learning-augmented SLAM.
- **Portability**: Easy integration into embedded platforms.
- **Foundation Layer**: Can be extended with inertial fusion to recover scale.
**Monocular SLAM Strategies**
**Feature-Based Methods**:
- Track sparse keypoints and build map landmarks.
- Robust and interpretable.
**Direct Methods**:
- Optimize photometric error over image intensities.
- Dense usage of image information.
**Visual-Inertial Extensions**:
- Add IMU to resolve scale and improve robustness.
- Common in mobile and drone systems.
**How It Works**
**Step 1**:
- Track visual correspondences and estimate relative camera motion.
**Step 2**:
- Triangulate landmarks, optimize local map, and apply loop closure for drift correction.
Monocular SLAM is **the most accessible SLAM configuration that delivers real-time mapping from a single camera while trading off direct metric scale observability** - with good initialization and optimization, it performs remarkably well in many settings.